--- base_model: krea/Krea-2-Turbo base_model_relation: adapter license: other license_name: krea-2-community-license license_link: https://huggingface.co/lvladikov/Krea2-Turbo-Distill-2step-LoRA/blob/main/LICENSE.pdf library_name: diffusers tags: - lora - text-to-image - distillation - step-distillation - distribution-matching - krea-2 pipeline_tag: text-to-image --- # Krea 2 Turbo β€” 2-Step Distillation LoRA > πŸ§ͺ **A fast-preview adapter, from a project still in training.** When the subject is close and fills a good part of the > frame β€” a portrait, a single figure, an object seen up close β€” two steps already hold up well, and you can rely on > this adapter for those images. Small subjects are where it still falls short, people and objects alike: faces in a > crowd, figures in a wide scene, the machines at the back of a gym β€” anything that takes up little of the frame can > come out ghosted or smeared. For those, and whenever quality matters more than speed, use the > [4-step LoRA](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA). Every figure on this page measures the > adapter honestly against 4-step and 8-step renders. Training continues one recipe change at a time, and a later > checkpoint replaces this one only when the sweeps and I visually agree it is better. Known issues β€” see > [Known issues](#known-issues). > > πŸ“ **The saved steps can also go into resolution.** A small subject is simply one that covers few pixels, so a larger > render makes the same subject bigger β€” and at a quarter of the teacher's steps, renders up to 2048Γ—2048, Krea's > published maximum recommended resolution and beyond the largest size this adapter was trained at (1440Γ—1440), come > within easy reach. That makes the adapter a stepping stone to high-resolution renders as well as a fast preview. Past > 2048Γ—2048, stock Krea 2 itself begins to duplicate subjects β€” a property of the base model, with or without this > adapter. A LoRA for **[Krea 2 Turbo](https://huggingface.co/krea/Krea-2-Turbo)** that takes the model from its usual **8 steps down to 2** β€” Turbo's own weights and its own two sigmas, guidance 0.0, a quarter of the denoising passes β€” aiming at the best quality two steps can give. Two steps give up more than four: this adapter is for **fast previews and drafts** at half the 4-step adapter's cost and a quarter of the teacher's, and the **[4-step LoRA](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA)** remains the recommendation for quality renders. - 🎯 **The aim** β€” the best two-step quality this base can give, at every one of the same 12 resolutions, measured against the 8-step teacher and against the [4-step LoRA](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA) as the reference. Not a claim to reach either. - ⚑ **A quarter of the steps** β€” 8 β†’ 2, on Turbo's own deployment sigmas. - ⏱️ **4Γ— faster denoising** β€” the model runs twice instead of eight times, and denoising is the part this adapter changes: **76.4 s β†’ 19.3 s** measured at 1024Γ—1024 on the same prompts, the adapter's own cost per call within measurement noise. What a whole render costs on top of that is unchanged by the LoRA and depends on your pipeline; see [Performance](#performance). - πŸ“Š **Distribution matching, not imitation** β€” the training objective that got the renders improving again after the 4-step project's recipe had stopped helping at two steps (see [Method](#method)). - πŸ—£οΈ **Prompt-conditioned throughout** β€” both scores in the distribution match, the teacher's and the fake adapter's, are evaluated on each prompt's own conditioning, so the student is matched to what the teacher makes _for that prompt_, not to a prompt-free look. There is no separate adherence term: instead a vision-language judge checks every checkpoint β€” each render scored alone against the prompt's objects, counts, attributes and relations, with the teacher scored the same way β€” and a term would only be added if that meter showed adherence slipping. The one part of training that looks at images without their prompt is the artefact critic (see [Method](#method)), and it only judges whether fine structure looks like the teacher's. - πŸ“ **12 trained resolutions** β€” multi-aspect from 512Γ—512 up to 1440Γ—1440, each with its sweep. - πŸ”Œ **Drop-in, no exceptions** β€” a plain LoRA sampled by stock Euler at sigmas `[1.0, 0.7595]` in diffusers, ComfyUI or MLX. No custom sampler, no policy head, no per-step tricks. If the quality needs a special sampler it is not this project. - 🧬 **Same shape as the [4-step adapter](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA)** β€” rank 64 on the same 228 modules; a second adapter exists during training only and never ships. - 🎲 **The same 13,750 recorded teacher trajectories** the [4-step adapter](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA) trained on, reused without a single teacher re-run. - πŸ”’ **41,320 training samples** in the 2-step stages, on top of the [4-step LoRA](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA)'s 78,000 β€” all of them drawn from the **same recorded material**: no new prompts, no new text embeddings and not one new teacher run. A training sample is one pass over a prompt that was already encoded and already traced by the teacher for the 4-step project, read again at the two sigmas this schedule uses. - πŸ“… **25 days** from the first 2-step training launch to this checkpoint, on a single RTX 3090 β€” and the project continues. - πŸ” **More than forty recipe adjustments** across two methods so far β€” seven of trajectory distillation before the switch, the rest of distribution matching since β€” each kept only when the renders did not get worse. - πŸ–₯️ **One RTX 3090**, and a recipe shaped by its 24 GB. [![The 22 test prompts, rendered by Krea 2 Turbo with this LoRA at 2 steps](assets/thumbs/poster.jpg)](assets/poster.jpg) _All of the above were created with this LoRA at 2 steps: the 22 test prompts, Krea 2 Turbo + the LoRA, seed 4242, each at one of its trained resolutions. Click for full size. The side-by-side comparisons with the 8-step teacher are in [Examples](#examples)._ ## Files | file | what it is | | -------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `krea2_turbo_2step_rank_64_lora.safetensors` | the LoRA in diffusers key format β€” see [Inference with diffusers](#inference-with-diffusers); also for MLX or anything that reads safetensors | | `krea2_turbo_2step_rank_64_lora_comfyui.safetensors` | the same weights under ComfyUI's key names β€” see [ComfyUI](#comfyui) | | `krea2_turbo_2step_lora_t2i.json` | a ready ComfyUI workflow, stock nodes only | | [`krea2_turbo_2step_rank_64_lora_checkpoint_info.md`](krea2_turbo_2step_rank_64_lora_checkpoint_info.md) | **the quick place to check which checkpoint the two weight files are based on.** The pair above keeps its names and is updated in place as better checkpoints ship; this file always says what they are today. Every published checkpoint also sits in [`_archive/checkpoints/`](_archive/checkpoints) under its number | | `LICENSE.pdf` | the Krea 2 Community License Agreement, which covers this adapter β€” see [License](#license) | | `NOTICE.txt` | the attribution notice the license requires of a derivative | The two weight files are one adapter β€” only the key names differ. Both carry the training details in their safetensors metadata: base model, lineage, method, the checkpoint and the inference settings. Their file names never change; when a better checkpoint ships they are replaced in place, and [`krea2_turbo_2step_rank_64_lora_checkpoint_info.md`](krea2_turbo_2step_rank_64_lora_checkpoint_info.md) is the quick place to check which checkpoint the current files are based on. ## Where it stands | | | | ------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | lineage | [4-step LoRA](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA) β†’ 2-step trajectory distillation β†’ distribution matching β†’ a spectral match against the teacher's own images β†’ an artefact critic and detail terms β†’ three critics taking turns β†’ nine critics, detail and colour held to the teacher region by region, a guarded running average β†’ the teacher's own layout held at the large sizes, eight critics, every update to the average checked for doubled subjects | | this release | the current run's latest probed checkpoint, chosen by the 12-bucket sweep and by my own look at the renders; the run continues from it one recipe change at a time | | what it gives | usable two-step renders at every trained resolution: fine detail and colour at the teacher's level β€” from 1 megapixel up, closer to the teacher than the 4-step adapter β€” with the prompt's objects, counts, attributes and relations in place (a blind rubric finds 2 points missing out of 352, the teacher's own renders 1). A judge asked which render follows the prompt better still prefers the 8-step teacher on 12 of 66 (the 4-step adapter: 6 of 45, on the original 15 prompts), mostly on how a stylised prompt says things should look. What it does not give is the teacher's own picture: see [Known issues](#known-issues) and [Measured against the teacher](#measured-against-the-teacher) | ### Known issues The usual costs of two steps, in this order of how often they show. **Small subjects are the weak spot, people and objects alike**: a portrait-sized face or an object seen up close holds up, while small or distant subjects β€” faces in a crowd, a figure in a wide scene, the machines at the back of a room β€” can come out ghosted, smeared or misshapen, since at that size a whole subject is only a few of the blocks the model works in. Fine structure can come out soft or a few pixels out of register β€” feathers, hair strands, signage, the surface of a distant object β€” most at 1280Γ—1280 and above, and a faint doubled contour can show on limbs. On stylised prompts, **how the prompt says the image should look is followed less faithfully than what should be in it**: crisp anime linework, energetic brush strokes, the fingerprints in clay or a matte-painting finish come out closer to a generic rendering than the teacher's. On busy action or crowd scenes the composition can repeat itself β€” an extra hand or held object, a figure duplicated in a crowd β€” where the 8-step and 4-step renders commit to one. On some prompts the composition itself differs from the 8-step render at the same seed: two steps is a shorter path from the same starting noise, so the image can settle on a different framing, pose or arrangement rather than a degraded version of the teacher's. Treat the teacher's render as a reference for quality, not as the picture two steps will reproduce. Skin reads smoother than the teacher's β€” its finest texture, the pores, sits under the teacher's at the larger sizes, though its colour is now close to the teacher's β€” and freckles come out as dots, softer than the teacher's. Every one of these is being worked on; none is hidden in the sweeps or the examples. ## How I got here The [4-step adapter](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA) closed its page with a promise: a 2-step LoRA as the next project, and a guess at the lever it would need β€” matching the teacher's _distribution_ rather than its trajectory. That guess turned out to be the whole story. The project began where the [4-step one](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA) ended, from its final weights, and ran the same recipe at two steps: **progressive distillation** on the recorded teacher trajectories, each student call covering four teacher steps, with the LADD-style critic as the finisher. Well into that run, every number had stopped moving and the pictures had a signature the numbers could not see: doubled contours on faces and limbs, soft fine texture, crowds averaged into translucent overlaps. Several variations followed β€” the critic re-weighted, judged per token, a heavier hand on the final call, the student's own first-step output fed into its second β€” and each traded one of those faults for another without moving past them. A capacity probe ruled out adapter rank; a learning-rate shock ruled out the optimiser. The reason is structural, and worth stating plainly because it decides the whole design. A regression loss asks the student to land on the teacher's _specific_ image for each prompt. When a two-step jump is wide enough that several images are plausible, the answer that minimises the squared error is their average β€” and the average of two sharp images is a blurred one with doubled edges. Every earlier recipe rewarded that average. Tuning its weights could not change what it rewarded. **Distribution matching** asks a different question: not "does your image match this one" but "would the teacher plausibly have produced your image". The first run of that objective, on top of the trajectory-distilled weights, produced in a fraction of the old recipe's training what all of it never had β€” and it did so while every latent distance to the teacher _rose_, which is exactly what a mode-seeking objective predicts and what a mean-seeking metric punishes. The distances are reported on this page; they are not optimised for, and they are not what decides a checkpoint. Pictures are, at fixed seeds, at every resolution, with faces viewed at 1:1. The recipe adjustments so far, each made on the measurement of the one before: 1. progressive distillation at two steps from the [4-step adapter](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA)'s weights, with the LADD critic β€” the baseline 2. the critic made to judge the first call's endpoint against the teacher's mid-states, which removed the gross ghosting and left the doubled contours 3. the critic's weight, its per-token form and the student's own first-step output as the second call's input β€” tried one at a time; the objective flip that mattered was not among them 4. **distribution matching** (the DMD2 family) as the primary objective, trajectory regression demoted to an anchor at half weight, the critic switched off to read the new term alone 5. the running average of the weights restarted at the objective switch, after it was caught averaging two lineages into composites 6. a spectral match of the student's image against the teacher's own, on the whole latent and on decoded pixel windows β€” the term that finally reached the grain at large resolutions, after a critic, per-resolution weights and a filtered push had each been tried against it and retired 7. the decoded-window spectral term given a much lighter hand β€” capped at a quarter of its earlier strength, which kept the detail and took some grain out of flat areas 8. the fake-score adapter updated four times per student step instead of twice, so it keeps up with what the student currently makes; faint straight-line artefacts that had begun to appear on flat illustrated areas went away with it 9. **an artefact critic** β€” a small head reading the frozen base model's own mid-network features, trained to tell the teacher's finished images from the student's, with its push restricted to structure finer than 32 pixels and held well below the distribution term β€” aimed at what the distribution term leaves behind: melted small faces and dense detail smeared into blotches 10. four detail terms together: the anchor counting the fine-detail part of its error twice; a one-sided floor that stops the finest detail dropping below the level real photographs carry; a second-call target made by the teacher finishing the image from the student's own first-call output, so the target shares the student's layout; and a smoothness limit on the fake-score adapter, so the distribution push keeps pointing at detail rather than away from it 11. **three critics taking turns** β€” the artefact critic joined by one whose real examples are half real photographs and one weighted toward faces, one of them pushing on each step while the others keep training in between, because the three did not fit in memory side by side 12. **a colour band** β€” a floor under the saturation of the whole image and of the decoded window at the teacher's own level, and a ceiling 8% above it, so colour can neither fall under the teacher's at large sizes nor climb past it; grey and black-and-white references are left alone 13. **more critics, each with one job** β€” nine in all, still taking turns: the photo critic confined to photographic prompts; a critic for large faces that pushes on every step, with the face critic kept beside it for photographs; and five new ones β€” the first call's layout judged against the teacher's own intermediate state from the same noise, content weighted toward wherever the teacher put detail, the prompt (the critic sees it, and is shown a mismatched one as a negative), text, and the teacher's own finish of the student's second-call state 14. **detail held to the teacher region by region** β€” the decoded-window spectral pull aimed at the teacher's most detailed tenth of the image, with the anchor counting fine error there three and a half times; a per-area ceiling at 1.3Γ— the teacher's detail; a direction-aware spectral term; on photographs, half the decoded windows centred on eyes, nose or lips; the photo floor confined to photographic prompts, with floors at the teacher's own level for stylised prompts and for flat areas 15. the second call's losses also reaching the first call, at a fifth of their strength, so the first call is shaped for the finish it feeds; small faces weighted up in the distribution term; and the fake-score adapter back to three updates per student step, its lag behind the student measured inside the range it had at four, which bought back training pace 16. **more training at large sizes and on stylised prompts** β€” 1440Γ—1440 from under 2% of the samples to about 6%, 1024Γ—1024 and 768Γ—1024 to 12% each; stylised prompts from 4% to 12% 17. **a guarded running average** β€” the published weights are a running average of training, and averaging two layouts of the same prompt had produced doubled subjects: a short-lived excursion of the training weights is now kept out of the average and a lasting change taken in whole, and the average was restarted once, after a layout change it had blended 18. **three of the newest additions taken back out** β€” the floors at the teacher's level for stylised prompts and for flat areas, and a face critic for photographs, all of which had trained the last stretch of the previous checkpoint: a render of the training weights drew one animal as two joined bodies, and the recipe went back to the one that had trained the weeks before, with eight critics 19. **a stricter guard on the running average** β€” a change in the training weights is taken into the average only once it has held for six rounds instead of three; shorter swings are left out 20. **the second call held to the teacher's layout** β€” the anchor to the teacher's recorded trajectory counts the coarse layout of the second call, everything 64 pixels and up, twice 21. **every update to the running average checked for doubled subjects first** β€” before a stretch of training goes into the average, the average it would make renders a fixed probe at the large sizes, and a stretch that would put a second head on the subject is held out 22. **the distribution term halved at the large sizes** β€” at every size with a 1280- or 1440-pixel side, while the anchor to the teacher's trajectory keeps its full weight there, so at those sizes the teacher's own layout carries twice the share it did; the doubled subjects had formed at exactly the sizes trained most 23. the running average rebuilt once more, from the training weights since the halving, after the guard had held it still through a long change of pose Two further ideas were tried and taken back out: confining the distribution term to the second call's noise range, and a detail pyramid compared pixel by pixel against the teacher, which on inspection rewarded fading any detail it could not place exactly where the teacher had it. ## chk00041320 vs chk00031600 `chk00031600` (published 25 Sep 2026) was the first checkpoint with a colour band, nine critics and a guarded running average. `chk00041320` (2 Oct 2026) is 9,720 training samples later, and those samples went to what matters first in a picture β€” subjects drawn twice, or two poses blended into one, at the large sizes β€” and to faces in busy scenes, through the changes numbered 18 to 23 under [How I got here](#how-i-got-here), each kept only after its own look at the renders: 1. three of the newest additions taken back out, eight critics 2. a stricter guard on the running average 3. the second call held to the teacher's layout 4. every update to the running average checked for doubled subjects 5. the distribution term halved at the large sizes 6. the running average rebuilt from the recent training It is also the first checkpoint measured on the full 22-prompt sweep: the original 15, plus seven scenes of people, animals and action added on 29 Sep 2026 because structure is what they test β€” five friends on a beach, a family at a table seen from above, three kittens in a basket, two dogs in a tug of war, a show jumper, a pianist's hands, a flock of flamingos. Every figure below compares the two checkpoints on those 22, each against the 8-step teacher. **Structure β€” the headline.** A judge compares each of the 21 structure renders β€” the seven scenes at 1280Γ—1280, 1440Γ—1440 and 1440Γ—1280 β€” with the teacher's for missing, extra or merged body parts and subjects, and every flag is checked by eye at 1:1. The previous checkpoint has one real fault among them β€” a ghost saddle pad and boot behind the show jumper's neck at 1280Γ—1280 β€” and this one has none, and the snow leopard the guard watches has come out as one animal at every large size in every checkpoint since the average was rebuilt. Doubling at the level of fine structure falls too: the two ghosting indexes come closer to the teacher at 11 and 10 of the 12 sizes, most at the large ones β€” at 1440Γ—1440 from 1.17Γ— the teacher's to 1.06Γ—. **Faces in busy scenes.** The market scene at 1280Γ—1280 β€” vendors leaning over a stall β€” now comes out with whole faces where the previous checkpoint drew a broken one, and its layout sits much closer to the teacher's (a correlation of the stall side with the teacher's render: 0.71 β†’ 0.79). Inside the faces the sweep detects, skin texture comes closer to the teacher's at 1280Γ—1280 and 1440Γ—1440 (0.89Γ— β†’ 0.90Γ—, 0.83Γ— β†’ 0.87Γ—), and skin colour at 1280Γ—1280 rises from 0.86Γ— of the teacher's to 0.95Γ—. **Detail and colour.** At the two largest sizes the excess energy two steps put into the grid bands comes down toward the teacher's level: | 1.00 = the teacher | fine texture | 16-px band | 8-px band | | --- | --- | --- | --- | | 1440Γ—1280 | 1.01 β†’ 0.99 | 1.03 β†’ **1.00** | 1.10 β†’ **1.09** | | 1440Γ—1440 | 1.07 β†’ **1.03** | 1.08 β†’ **1.05** | 1.13 β†’ **1.09** | At 1280Γ—1280 and below the fine texture stays within a few percent of the teacher's, and the grain in flat areas β€” skies, walls, out-of-focus backgrounds β€” comes closer to the teacher's at 8 of the 12 sizes: 1.03Γ— the teacher's across the sweep, from 1.07Γ—. Edge detail rises at 768Γ—1024 and 1280Γ—1280 (0.91Γ— β†’ 0.94Γ—, 0.91Γ— β†’ 0.93Γ—), and colour stays at the teacher's level: 0.99Γ— across the sweep, from 0.98Γ—. **Prompt adherence.** The judge that asks which of two renders follows the prompt better prefers the teacher on 12 of 66 β€” the 22 prompts at 512Γ—512, 1280Γ—1280 and 1440Γ—1440 β€” against 15 for the previous checkpoint, and gives this one the only outright win. The blind rubric, now 22 prompts at four sizes, finds the same 2 points missing out of 352 for both; the teacher's own renders lose 1. On 15 prompts drawn fresh from the prompt bank for this checkpoint, plus 5 black-and-white ones: 0 wins, 14 ties, 1 loss, and 0 Β· 4 Β· 1 in black and white. **Speed β€” unchanged.** The same adapter shape at the same cost: measured again at 1024Γ—1024, two steps with this LoRA took 19.3 and 19.2 s against 76.4 and 76.3 s for the teacher's eight β€” 4.0Γ—, two model calls instead of eight. **What stays a limit of two steps.** Small subjects in wide scenes and the tactile surface of stylised materials such as clay stay where two steps fall furthest short of eight. The figures are in [Measured against the teacher](#measured-against-the-teacher). | axis | `chk00031600` | `chk00041320` | | --- | --- | --- | | structure renders with a fault the teacher's does not have (of 21, checked at 1:1) | 1 | **0** | | ghosting, blur-invariant index, sweep mean | 1.24Γ— | **1.21Γ—** | | market faces at 1280Β², layout closeness to the teacher | 0.71 | **0.79** | | grain in flat areas, sweep mean | 1.07Γ— | **1.03Γ—** | | fine texture vs the teacher, 1440Β² | 1.07 | **1.03** | | judge prefers the teacher (of 66) | 15 | **12** | | blind adherence rubric, points missing of 352 | 2 | 2 | | saturation vs the teacher, sweep mean | 0.98Γ— | **0.99Γ—** | | distance to the teacher, sweep mean | 0.41 | 0.41 | | training samples in the 2-step stages | 31,600 | 41,320 | ## chk00031600 vs chk00017464 `chk00017464` (published 14 Sep 2026) was the first checkpoint with critics and detail terms. `chk00031600` (25 Sep 2026) is 14,136 training samples later, and those samples went to the faults its page listed β€” grain and excess texture at the largest sizes, colour running under the teacher's there, small faces, and stylised prompts drifting toward a generic look β€” through the changes numbered 12 to 17 under [How I got here](#how-i-got-here), each kept only after its own look at the renders: 1. a colour band at the teacher's level 2. nine critics taking turns, each with one job 3. detail held to the teacher region by region, with a ceiling as well as floors 4. the second call's losses reaching the first call, and small faces weighted up in the distribution term 5. more training at the large sizes and on stylised prompts 6. a guarded running average **Grain and texture at large sizes β€” the headline.** Every one of the 12 sweep resolutions Γ— 15 prompts measured against the 8-step teacher, as in [Measured against the teacher](#measured-against-the-teacher). The excess fine energy two steps used to put into large images is gone: fine texture and both grid bands sit within 5% of the teacher's at 1280Γ—1280 and 1440Γ—1280, and within 13% at 1440Γ—1440: | 1.00 = the teacher | fine texture | 16-px band | 8-px band | | --- | --- | --- | --- | | 1280Γ—1280 | 1.09 β†’ **0.95** | 1.08 β†’ **0.95** | 1.13 β†’ **0.96** | | 1440Γ—1280 | 1.08 β†’ **0.95** | 1.09 β†’ **0.97** | 1.14 β†’ **0.99** | | 1440Γ—1440 | 1.17 β†’ **1.08** | 1.08 β†’ **1.05** | 1.21 β†’ **1.13** | The 8-px band comes closer to the teacher at all 12 resolutions, the 16-px band at 9 and fine texture at 7, and the grain in flat areas β€” skies, walls, out-of-focus backgrounds β€” at 10 of 12: 1.08Γ— the teacher's across the sweep, from 1.26Γ— (1440Γ—1440: 1.57Γ— β†’ 1.14Γ—, median of the 15 prompts). The two ghosting indexes are about level (closer at 7 and at 5 of the 12), and the distance to the teacher β€” a plain pixel measure this page reports but training does not optimise β€” is within a hundredth of the previous checkpoint's. **Colour, now at the teacher's level.** Saturation across the sweep rises to 0.98Γ— the teacher's from 0.93Γ—, closer at 10 of the 12 sizes; at 1440Γ—1440 it goes from 0.86Γ— to 0.96Γ—, at 1280Γ—1280 from 0.87Γ— to 0.93Γ—. Skin inside detected faces follows: its saturation from 0.92Γ— to 0.97Γ— at 768Γ—1024 and from 0.85Γ— to 0.94Γ— at 1440Γ—1440. **Small faces.** A crowd probe β€” 15 prompts full of small faces at 1280Γ—1280 β€” checks every frontal face for eyes, nose and mouth in place, with the check calibrated on the teacher's own faces. **63% of the faces keep that structure, up from 53%, against the teacher's 64%**; the blur that brings the teacher's own faces down to the same pass rate falls from 1.7 pixels to 1.0. The gain is largest on faces 32–48 pixels tall (56% β†’ 69%) and 64–96 pixels tall (73% β†’ 87%). Small subjects stay the part of the image two steps find hardest β€” see [Known issues](#known-issues). **Faces up close.** On the test portrait at 1:1 the freckles come out as separate dots rather than the clusters of the previous checkpoint, and the eyes stay clean, irises and catchlights in place; the skin between the freckles is smoother than the teacher's. **Prompt adherence β€” held.** The judge that asks which of two renders follows the prompt better prefers the teacher on 11 of 45, as before; the blind rubric finds the same 1 point missing out of 240; and on 15 prompts drawn fresh from the prompt bank for this checkpoint, plus 5 black-and-white ones, both checkpoints come out the same: 1 win, 9 ties, 5 losses, and 0 Β· 4 Β· 1 in black and white. **Speed β€” unchanged.** The same adapter shape at the same cost: measured again at 1024Γ—1024, two steps with this LoRA took 19.8 and 20.2 s against 84.3 s for the teacher's eight β€” the same 4.2Γ—. **What stays a limit of two steps.** Small subjects in wide scenes, and the tactile surface of stylised materials such as clay, are still where two steps fall furthest short of eight. Both were worked on across these samples β€” face critics of several kinds, a small-face curriculum, small faces weighted up in the distribution term, a critic on the teacher's own finish β€” and both remain the focus of what comes next. Fine edges and skin texture at the largest sizes sit a little under the teacher's; the figures are in [Measured against the teacher](#measured-against-the-teacher). | axis | `chk00017464` | `chk00031600` | | --- | --- | --- | | fine texture vs the teacher, 1280Β² / 1440Β² | 1.09 / 1.17 | **0.95 / 1.08** | | 16-px grid band, 1280Β² / 1440Β² | 1.08 / 1.08 | **0.95 / 1.05** | | grain in flat areas, sweep median | 1.26Γ— | **1.08Γ—** | | saturation vs the teacher, sweep mean | 0.93Γ— | **0.98Γ—** | | small faces keeping their structure (the teacher: 64%) | 53% | **63%** | | judge prefers the teacher (of 45) | 11 | 11 | | blind adherence rubric, points missing of 240 | 1 | 1 | | distance to the teacher, sweep mean | **0.406** | 0.413 | | training samples in the 2-step stages | 17,464 | 31,600 | ## chk00017464 vs chk00013663 `chk00013663` (published 12 Sep 2026) was the first public checkpoint: distribution matching with the spectral match, and nothing yet aimed at the faults the distribution term leaves behind. `chk00017464` (14 Sep 2026) is 3,801 training samples later, and every one of those samples went to those faults β€” the grain and grid pattern at large sizes, small faces, dense detail β€” through five recipe changes, each kept only after its own look at the renders: 1. the decoded-window spectral term brought down to a quarter of its strength 2. the fake-score adapter updated four times per student step instead of twice 3. the artefact critic, reading the frozen base model's own features 4. four detail terms: the anchor counting fine-detail error twice, the one-sided photo floor, the teacher's finish of the student's first call as the second call's target, and a smoothness limit on the fake adapter 5. three critics taking turns: the artefact critic, a photo critic and a face critic Two ideas were tried and taken back out along the way: the distribution term confined to the second call's noise range, and a detail pyramid compared pixel by pixel against the teacher. **Detail at large sizes β€” the headline.** Every one of the 12 sweep resolutions Γ— 15 prompts measured against the 8-step teacher, as in [Measured against the teacher](#measured-against-the-teacher). Above 1 megapixel the excess fine energy two steps used to put into images is roughly halved, and those sizes are now closer to the teacher than the [4-step adapter](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA) on fine texture and on both grid bands: | 1.00 = the teacher | fine texture | 16-px band | 8-px band | | --- | --- | --- | --- | | 1280Γ—1280 | 1.20 β†’ **1.09** | 1.16 β†’ **1.08** | 1.25 β†’ **1.13** | | 1440Γ—1280 | 1.21 β†’ **1.08** | 1.19 β†’ **1.09** | 1.28 β†’ **1.14** | | 1440Γ—1440 | 1.34 β†’ **1.17** | 1.25 β†’ **1.08** | 1.42 β†’ **1.21** | Fine-texture energy and both grid bands come closer to the teacher at 10 of the 12 resolutions, the blur-invariant ghosting index at 10 of 12, the micro-ghost index at all 12, and the grain in flat areas β€” skies, walls, out-of-focus backgrounds β€” at 11 of 12 (1440Γ—1440: 1.84Γ— the teacher's β†’ 1.57Γ—, median of the 15 prompts). It shows where it should: hair renders as more individual strands, and the token-grid texture that read as grain on birds, food and foliage at 1440Γ—1440 is lighter. **Faces and skin.** Freckles on the test portrait gather into lighter, more dot-like clusters than before β€” better, not yet the teacher's separate dots β€” and eyes look about the same: clean irises, lashes softer than the teacher's. **What did not improve.** The judge that asks which of two renders follows the prompt better prefers the 8-step teacher on 11 of 45 against 6 before, mostly on how a stylised prompt says the picture should look: crisp linework, energetic brush strokes, the texture of clay. The blind rubric that checks each render alone for the prompt's objects, counts, attributes and relations moves from 0 to 1 point missing out of 240, and on the same 15 fresh prompts the previous checkpoint was measured on, the two checkpoints come out level (1 win, 13 ties, 1 loss). Colour runs slightly lower (saturation 0.93Γ— the teacher's across the sweep, from 0.94Γ—; 0.86Γ— at 1440Γ—1440), and at 512Γ—512 the two checkpoints are level. Both are what the next recipe changes target. | axis | `chk00013663` | `chk00017464` | | --- | --- | --- | | fine texture vs the teacher, 1280Β² / 1440Β² | 1.20 / 1.34 | **1.09 / 1.17** | | 16-px grid band, 1280Β² / 1440Β² | 1.16 / 1.25 | **1.08 / 1.08** | | grain in flat areas, sweep median | 1.36Γ— | **1.26Γ—** | | distance to the teacher, sweep mean | 0.413 | **0.406** | | judge prefers the teacher (of 45) | **6** | 11 | | blind adherence rubric, points missing of 240 | **0** | 1 | | saturation vs the teacher, sweep mean | **0.94Γ—** | 0.93Γ— | | training samples in the 2-step stages | 13,663 | 17,464 | ## Measured against the teacher Every number here compares a render of this LoRA at 2 steps with the 8-step reference render of the **same prompt at the same seed**, across the 22 test prompts and the 12 trained resolutions. Two other columns are measured the same way, so the figures have a floor and a ceiling around them: **stock Krea 2 Turbo at 2 steps**, which is what the base model does without the adapter, and the [4-step LoRA](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA) at its own 4 steps, which is the better tool and the thing worth being compared against β€” its renders exist for the original 15 prompts, so its figures are on those 15. **Detail, band by band.** Fine-texture energy and the two grid bands, as a ratio to the teacher's own (1.00 = the teacher): | bucket | fine texture | 16-px band | 8-px band | 4-step LoRA (same three, original 15) | stock 2-step, fine texture | | --- | --- | --- | --- | --- | --- | | 512Γ—512 | 1.00 | 1.01 | 1.09 | 1.05 Β· 1.05 Β· 1.07 | 0.65 | | 768Γ—1024 | 1.00 | 0.98 | 1.02 | 1.05 Β· 1.05 Β· 1.08 | 0.44 | | 1024Γ—1024 | 1.06 | 1.10 | 1.13 | 1.11 Β· 1.17 Β· 1.16 | 0.42 | | 1280Γ—1280 | 1.02 | 1.02 | 1.03 | 1.20 Β· 1.18 Β· 1.22 | 0.44 | | 1440Γ—1440 | 1.03 | 1.05 | 1.09 | 1.24 Β· 1.23 Β· 1.32 | 0.52 | Two steps without the adapter carry 0.65Γ— the teacher's fine detail at 512Γ—512 and **about half or less** (0.42–0.52Γ—) at the four larger sizes. With it, the fine texture sits at the teacher's level everywhere β€” between 0.97Γ— and 1.09Γ— of it across all 12 sizes β€” and from 1 megapixel up it is closer to the teacher than the 4-step adapter, which carries more excess fine energy there. **Prompt adherence, judged.** A vision-language judge is shown the teacher's render and this LoRA's for the same prompt, in both orders, and asked which follows the prompt better; a loss means the teacher was preferred both times: | bucket | wins | ties | losses | | --- | --- | --- | --- | | 512Γ—512 | 1 | 15 | 6 | | 1280Γ—1280 | 0 | 18 | 4 | | 1440Γ—1440 | 0 | 20 | 2 | Twelve losses out of 66 (the [4-step LoRA](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA) scores six of 45 against the same teacher on the original 15). They gather on stylised prompts and on how a prompt says the picture should look β€” crisp linework, energetic brush strokes, the texture of clay β€” more than on what should be in it: a blind rubric that scores each render on its own against the prompt's objects, counts, attributes and relations, with the teacher scored identically, finds 2 points missing out of 352 (the teacher's own renders lose 1). On 15 prompts drawn fresh from the training prompt bank for this checkpoint and never rendered before, the judge returned 0 wins, 14 ties and 1 loss, and on five fresh black-and-white prompts 0 wins, 4 ties, 1 loss β€” with no colour cast in any of the five. **Checked for the damage this kind of training can do.** Saturation sits at 1.04Γ— the teacher's at 768Γ—1024 and 0.98Γ— at the larger sizes; edge detail 0.92–0.94Γ—; skin texture inside detected faces 1.02Γ— at 768Γ—1024 and 0.90Γ— / 0.87Γ— at 1280Γ—1280 / 1440Γ—1440, with skin saturation 0.98Γ— and 0.95Γ— at the larger sizes. The honest reading: **colour is at the teacher's level at every size, and skin remains the softest part of this adapter's output at large sizes.** Fine detail in flat regions β€” skies, walls, out-of-focus backgrounds β€” runs 1.32–1.57Γ— the teacher's on this measure, which is where two steps put grain that eight steps do not. **Distance to the teacher**, as a plain pixel measure, is 0.39–0.43 at every size against the [4-step LoRA](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA)'s 0.30–0.37 (on the original 15). That gap is what two model calls cost instead of four: the image is a good render of the prompt, but it is not the teacher's render of it β€” see [Known issues](#known-issues). ## Usage | setting | value | | -------------- | ----------------------------------------------- | | base model | Krea 2 Turbo | | LoRA scale | 1.0 | | steps | **2** | | guidance / CFG | **0.0** (Turbo is CFG-free; do not enable it) | | timestep shift | **mu = 1.15**, fixed (Turbo's deployment shift) | The 2 sampling sigmas are Turbo's own deployment grid: `[1.0, 0.7595]` β€” the first and the middle of the 4-step grid, so the model is evaluated at two points it already knows. ## Using it on Raw **This LoRA is trained on Krea 2 Turbo, against Turbo as its own teacher, and for Turbo.** Every layer it targets also exists in Krea 2 Raw, so it will load there without complaint β€” but that is a side effect of the shared architecture, not a supported mode: it is neither trained nor tuned for Raw's weights, steps or guidance. ## Inference with diffusers Krea 2 Turbo has a native diffusers pipeline, `Krea2Pipeline`, in diffusers from source β€” the same setup as Krea's own model card. The LoRA loads through that pipeline's standard LoRA loader, with one line of preparation: diffusers expects a `transformer.` prefix on every key, and it does not read the per-module `alpha` entries this file carries (alpha equals the rank, so dropping them changes nothing β€” the scale stays 1.0). Why the file is laid out the way it is, and what each runtime expects, is in [The keys, runtime by runtime](#the-keys-runtime-by-runtime). ``` pip install git+https://github.com/huggingface/diffusers.git ``` ```python import torch from diffusers import Krea2Pipeline from huggingface_hub import hf_hub_download from safetensors.torch import load_file pipe = Krea2Pipeline.from_pretrained("krea/Krea-2-Turbo", torch_dtype=torch.bfloat16).to("cuda") # "mps" on Apple Silicon lora = hf_hub_download("lvladikov/Krea2-Turbo-Distill-2step-LoRA", "krea2_turbo_2step_rank_64_lora.safetensors") state = load_file(lora) state = {f"transformer.{k}": v for k, v in state.items() if not k.endswith(".alpha")} pipe.load_lora_weights(state, adapter_name="2step") image = pipe("a fox in the snow", num_inference_steps=2, guidance_scale=0.0).images[0] image.save("krea2_2step.png") ``` - **`num_inference_steps=2` is the whole configuration.** The pipeline applies Turbo's fixed timestep shift (mu = 1.15) on its own and evaluates the model at Οƒ = 1.0 and 0.7595 β€” exactly the two points in [Usage](#usage) that the LoRA was trained on. Keep `guidance_scale=0.0`. - **Strength:** `pipe.set_adapters(["2step"], adapter_weights=[0.75])`. Stock Turbo is one call away for a side-by-side: `pipe.unload_lora_weights()` and `num_inference_steps=8`. - Use the diffusers file, not the `_comfyui` one: diffusers' Krea 2 key converter reads Krea's reference-trainer naming, not ComfyUI's `lora_down`/`lora_up`. ## ComfyUI A pre-converted file (`..._comfyui.safetensors`) and a ready workflow sit in the repo root. **No custom nodes** β€” stock ComfyUI only. [![ComfyUI workflow](assets/thumbs/workflow_preview.jpg)](assets/workflow_preview.jpg) | file | put it in | | ----------------------------------------------------------------------------------------------------------------------- | ---------------------------------- | | [`krea2_turbo_2step_rank_64_lora_comfyui.safetensors`](krea2_turbo_2step_rank_64_lora_comfyui.safetensors) | `ComfyUI/models/loras/` | | `krea2_turbo_bf16.safetensors` β€” [Comfy-Org/Krea-2](https://huggingface.co/Comfy-Org/Krea-2/tree/main/diffusion_models) | `ComfyUI/models/diffusion_models/` | | `qwen3vl_4b_bf16.safetensors` β€” [same repo](https://huggingface.co/Comfy-Org/Krea-2/tree/main/text_encoders) | `ComfyUI/models/text_encoders/` | | `qwen_image_vae.safetensors` β€” [same repo](https://huggingface.co/Comfy-Org/Krea-2/tree/main/vae) | `ComfyUI/models/vae/` | Then load [`krea2_turbo_2step_lora_t2i.json`](krea2_turbo_2step_lora_t2i.json). **The workflow is full bf16, with no quantisation anywhere.** bf16 needs no backend-specific kernel, so it runs unchanged on CUDA, Apple Silicon and CPU β€” one workflow, no platform caveats, nothing that depends on which device a component happens to land on. Smaller builds work too; both loaders accept any variant, just set the matching filename: | diffusion model | size | NVIDIA | Apple Silicon | | ----------------------------------------- | ------- | ------ | ------------- | | **`krea2_turbo_bf16`** (workflow default) | 26.3 GB | βœ… | βœ… | | `krea2_turbo_int8_convrot` | 13.5 GB | βœ… | βœ… | | `krea2_turbo_fp8_scaled` | 13.1 GB | βœ… | ❌ | The text encoder ships as bf16 (8.9 GB) or fp8 (5.2 GB) only β€” **there is no int8 text encoder**, so a fully matched int8 pair is not possible. **The LoRA is independent of the base build.** It is applied on top of the diffusion model by ComfyUI's own loader, which handles any dequantisation, so a quantised or otherwise optimised build of Krea 2 Turbo behaves just as bf16 does. Please use whichever variant suits your hardware β€” set it in the **Load Diffusion Model** node and leave the rest of the workflow untouched. The workflow ships bf16 simply because it is the one build guaranteed to run everywhere. > 🍎 **`fp8_scaled` does not work on Apple Silicon.** MPS has no `Float8_e4m3fn` support, so the run > dies at the sampler with _"Trying to convert Float8_e4m3fn to the MPS backend but it does not have > support for that dtype"_. That failure is the weight dtype, not the workflow or the LoRA β€” the > graph executes fine right up to the sampler. The fp8 _text encoder_ does run on MPS, but only > because ComfyUI places it on CPU; the workflow does not rely on that. > πŸ“Œ **The `_comfyui` file carries the same weights as the diffusers file** β€” only the key names differ. **Why a separate file.** ComfyUI addresses the transformer by its own layer names, so the adapter needs a key remap: `transformer_blocks.0.attn.to_q.lora_A` becomes `diffusion_model.blocks.0.attn.wq.lora_down`. **The tensors are bit-identical** β€” nothing is requantised or rescaled, only renamed. The mapping was verified against Comfy-Org's own Krea 2 LoRA: all 456 tensors land on keys that file also uses, with matching shapes. `alpha` keys are omitted, as in Comfy's own file. ComfyUI defaults alpha to the rank when absent, giving scale = alpha/rank = 1.0 β€” exactly what alpha 64 at rank 64 encodes. ### Settings | | | | ------------------- | ------------------ | | steps | **2** | | **cfg** | **1.0** | | sampler / scheduler | `euler` / `simple` | | LoRA strength | 1.0 | > βš™οΈ **`cfg 1.0`, not `0.0`.** ComfyUI expresses "no classifier-free guidance" as cfg **1.0**, whereas > diffusers expresses the same thing as guidance **0.0**. They mean the same: one forward pass per > step, no negative branch. Setting 0.0 in ComfyUI is not the same thing and will not give you > Turbo's intended behaviour. That is also why the workflow's negative input is a > `ConditioningZeroOut` β€” at cfg 1.0 it is never evaluated, so there is nothing to write in it. To compare against stock Turbo, set steps back to 8 and bypass the LoRA node with `Ctrl+B`. ## Performance Measured on this machine (Apple Silicon, MLX, bf16) at 1024Γ—1024, two prompts per configuration, each run on its own with the model already loaded, so the numbers are the render itself and not a model load: | configuration | denoising | per model call | GPU peak | | --- | --- | --- | --- | | Krea 2 Turbo β€” 8 steps (the reference) | **76.4 s** | 9.5 s | 25.2 GiB | | Krea 2 Turbo β€” 2 steps, no LoRA | 19.4 s | 9.7 s | 25.2 GiB | | **Krea 2 Turbo β€” 2 steps + this LoRA** | **19.3 s** | 9.6 s | 25.2 GiB | **Denoising is 4.0Γ— faster than the 8-step reference** β€” two model calls instead of eight. The adapter's own cost per call did not show up in this measurement: the runs with it came in marginally faster than those without, which is measurement noise, not a speed-up. A rank-64 low-rank product is small beside the transformer it is added to, and it adds no measurable memory. Denoising is the part the step count changes. What a complete render costs on top of it β€” encoding the prompt, decoding the latent, writing the file β€” is the same whether you run two steps or eight, and it depends on your pipeline, so the end-to-end figure on your machine will sit below 4Γ— and rise toward it as the render gets larger. **By resolution.** The two model calls of this LoRA's own sweep renders on the same machine (median of the 22 test prompts per size; sweep renders run one at a time, not the controlled measurement above, so a second or two either way between one sweep and the next is run-to-run variation β€” the adapter's shape, and so its cost, is the same at every checkpoint): | resolution | denoising (2 calls) | resolution | denoising (2 calls) | | --- | --- | --- | --- | | 512Γ—512 | 6.5 s | 1024Γ—1024 | 19.5 s | | 512Γ—768 / 768Γ—512 | 10.7 s / 11.3 s | 1280Γ—960 / 960Γ—1280 | 22.8 s / 22.7 s | | 768Γ—768 | 12.7 s | 1280Γ—1280 | 32.0 s | | 768Γ—1024 / 1024Γ—768 | 15.1 s / 15.2 s | 1440Γ—1280 / 1440Γ—1440 | 33.3 s / 38.3 s | ## LoRA strength Use **1.0**. That is the value the adapter was trained at, and where the 2-step output sits closest to the 8-step reference on every measurement in this page. [![LoRA strength comparison](assets/thumbs/portrait_strength_sweep.jpg)](assets/portrait_strength_sweep.jpg) The three panels individually: [0.5](assets/portrait_strength_0.5.jpg) Β· [1.0](assets/portrait_strength_1.0.jpg) Β· [1.5](assets/portrait_strength_1.5.jpg) β€” 1024Γ—1024, seed 4242, 2 steps. What the dial scales is this adapter's whole job at two steps: turning a pair of coarse calls into a finished image. So it behaves differently from a 4-step adapter's strength control, where the base render is already coherent and the LoRA only adds the missing texture. | strength | what happens | | --- | --- | | **below 1.0** | the correction is only partly applied β€” softer skin and hair, less fine structure, closer to what two steps look like without the adapter. There is less reason to reach for it here than at four steps, where the base render stands on its own | | **1.0** | the trained point, and the recommendation | | **1.0–1.5** | extrapolation past training: texture grows denser than the subject warrants and fine structure begins to read as wiry rather than sharp. Usable if you want that look, on a prompt-by-prompt basis | | **above 1.5** | not recommended, and not measured here | If a render is not giving you what you want, **reach for steps before strength**: this adapter is trained for two, and the [4-step adapter](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA) is the better tool whenever quality matters more than speed. ## File format and compatibility A plain `.safetensors` file β€” **not tied to any framework or backend**. It is weights plus a naming convention, so it loads under PyTorch (CUDA, MPS or CPU), MLX on Apple Silicon, or anything else that can read safetensors and do a matrix multiply. | | | | --------------- | --------------------------------------------- | | container | `safetensors` | | adapter weights | bf16 (`lora_A`, `lora_B`) | | `alpha` | fp32 scalar per module, **64.0** | | rank | 64 β†’ effective scale `alpha / rank` = **1.0** | Keys are **diffusers module paths** with PEFT-style suffixes: ``` transformer_blocks.0.attn.to_gate.lora_A.weight (64, 6144) transformer_blocks.0.attn.to_gate.lora_B.weight (6144, 64) transformer_blocks.0.attn.to_gate.alpha scalar time_embed.linear_2.lora_A.weight ... ``` applied the standard way: ``` W' = W + (alpha / rank) Β· (B @ A) ``` The one thing to watch when porting is **naming, not framework**. Runtimes that use their own layer names β€” ComfyUI, for instance, calls these `diffusion_model.blocks.N.attn.gate` with `lora_down`/`lora_up` β€” need a key remap first. The tensors themselves need no conversion. ### The keys, runtime by runtime The module paths in the file are the ones diffusers' `Krea2Transformer2DModel` uses for its layers β€” `transformer_blocks.N.attn.to_q`, `ff.up`, `time_embed.linear_2`, and so on β€” so they name the right tensors in any runtime that follows the diffusers architecture. What differs between runtimes is the wrapping around those paths: | runtime | what it expects | what to do | | ---------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------- | | **diffusers** (`pipe.load_lora_weights`) | every key prefixed with the pipeline component it belongs to β€” `transformer.` here β€” because one pipeline LoRA file may carry adapters for several components; the scale comes from the adapter config, so the loader drops `alpha` | add the prefix and drop `.alpha`, as in the snippet above | | **ComfyUI** | its own layer names, `diffusion_model.blocks.N.attn.wq` with `lora_down`/`lora_up`, no `alpha` keys (absent alpha defaults to the rank) | use the `_comfyui` file | | **MLX and custom loaders** | nothing in particular | read the keys as they are and apply `W + (alpha/rank)Β·(B @ A)` | In every case the tensors are the same 456 bf16 matrices; only the names around them change. ## Method **Distribution matching with a trajectory anchor**, Krea 2 Turbo as its own teacher, on the recorded 8-step trajectories. The student makes two calls, at Οƒ = 1.0 and Οƒ = 0.7595 β€” the first and fifth points of the teacher's 8-step grid at mu = 1.15 β€” and stock Euler carries it between them. That grid is what makes the objective a drop-in: Euler's first step from pure noise lands _exactly_ on the flow-matching interpolant at Οƒ = 0.7595 with the same noise and the student's own clean-image prediction as the data point. So the student's first-call output is a legitimate image prediction that can be judged as an image, and the second call is fed from it during training the way it will be at inference. **The distribution term.** For an image the student produces, two denoisers estimate how it should be cleaned up from a freshly noised copy: the frozen teacher, and a second small adapter on the same frozen base β€” the _fake score_ β€” that is trained online to denoise whatever the student currently makes. Where the two disagree is the direction that makes the image more like the teacher's work and less like the student's habits, and the student is pushed that way (the DMD2 gradient, per-sample normalised). Averaging is never rewarded, so the student commits. The fake adapter is rank 32, starts as an exact copy of the teacher, updates three times per student step β€” often enough to keep up with a student that is still changing β€” and is discarded at the end. On prompts with small faces, the push on the face tokens is weighted up, because small faces are what two steps get wrong most often. **The anchor.** Plain trajectory regression on the teacher's recorded chords stays in at half weight. It keeps the student on the teacher's two-step grid so the distribution term cannot wander into a different sampler behaviour, and it is what the earlier trajectory distillation had already satisfied β€” which is why the first distribution-matching run moved so far so fast. **Per resolution.** The distribution term's push grows with resolution: the fake adapter sees few large-bucket samples and under-fits fine structure there, and a per-pixel normaliser lands harder as pixel counts grow. A full 12-bucket sweep of the first distribution-matching checkpoint located the problem at 1 megapixel and above (fine-texture energy 1.4–1.8Γ— the teacher's at the five largest buckets); scaling the push per bucket from that measurement was tried and did not hold, and the spectral match replaced it. The same sweep of the published checkpoint, its running-average weights, fixed seed, 22 prompts per bucket, every image measured against the teacher's render of the same prompt and seed (in brackets: the first distribution-matching checkpoint, on the original 15 prompts): | bucket | fine texture vs the teacher | 16-px grid band | 8-px grid band | distance to the teacher | | --------- | --------------------------- | --------------- | -------------- | ----------------------- | | 512x512 | 1.00 (1.16) | 1.01 (1.23) | 1.09 (1.23) | 0.43 (0.44) | | 512x768 | 0.97 (1.23) | 0.99 (1.37) | 1.02 (1.33) | 0.41 (0.41) | | 768x512 | 1.01 (1.22) | 1.07 (1.37) | 1.02 (1.26) | 0.42 (0.47) | | 768x768 | 1.05 (1.43) | 1.07 (1.46) | 1.09 (1.50) | 0.40 (0.43) | | 768x1024 | 1.00 (1.36) | 0.98 (1.38) | 1.02 (1.42) | 0.39 (0.43) | | 1024x768 | 0.98 (1.35) | 0.98 (1.40) | 0.98 (1.38) | 0.40 (0.46) | | 1024x1024 | 1.06 (1.47) | 1.10 (1.55) | 1.13 (1.56) | 0.42 (0.44) | | 1280x960 | 0.97 (1.50) | 0.99 (1.59) | 0.98 (1.56) | 0.43 (0.42) | | 960x1280 | 1.09 (1.60) | 1.02 (1.63) | 1.13 (1.70) | 0.42 (0.43) | | 1280x1280 | 1.02 (1.65) | 1.02 (1.68) | 1.03 (1.69) | 0.43 (0.44) | | 1440x1280 | 0.99 (1.59) | 1.00 (1.66) | 1.09 (1.66) | 0.40 (0.42) | | 1440x1440 | 1.03 (1.81) | 1.05 (1.79) | 1.09 (1.89) | 0.40 (0.41) | ### The spectral match Distribution matching delivers its push one latent token at a time, and a token is a 16Γ—16-pixel block. At large resolutions the push that tells the student to add detail also, unintentionally, adds detail in the shape of that block: a faint checkerboard at the 16-pixel and 8-pixel periods that reads as grain on birds, food and foliage. It is measurable in the images' Fourier spectra β€” the student carried about twice the teacher's energy at exactly those two periods β€” and several ways of reshaping or re-weighting the push (a critic, per-region weights, a filtered push) were tried and retired: none of them could reach a pattern that lives inside the token. What can is a term that judges the **picture**, not the push. For every second-call sample, the student's estimate of the finished image and the teacher's own image for the same prompt and seed are compared through their radial power spectra β€” how much energy each has at every spatial scale β€” as an L1 on the log spectrum, in two places: on the whole latent every sample, and on one random 256-pixel window decoded through the VAE so colour and decoder behaviour are judged too. The comparison is two-sided, so too much fine energy and too little are both penalised; a blurred image does not satisfy it. Its gradient is added to the distribution push and capped per sample as a fraction of it, so it refines rather than takes over. At its first strength it brought every resolution closer to the teacher's spectrum without touching adherence, layout or variety; raised, it began closing the grain on the hardest subjects too. The whole-latent comparison trains at that strength; the decoded window was later brought down to a quarter of it, which kept the detail and removed some grain. Later still its pull was aimed: 0.4Γ— the distribution term on the tenth of the image where the teacher has the most fine detail, and 0.05Γ— everywhere else, so detail is matched where the teacher has it and flat areas are left alone. ### The artefact critic Distribution matching improves what the fake adapter can see, and the fake adapter learns from the student's own images β€” so where the student smears something, the fake learns the smear and the push stops correcting it. The two places that shows most are small faces, which come out melted, and dense content such as the goods on a market stall or the shelves of a shop seen through its window, which comes out as coloured blotches. A critic breaks that loop by looking at finished images instead. It is a small head on the frozen base model's mid-network features β€” every adapter switched off, the forward pass stopped halfway, so no adapter can learn to fool it and nothing it learns leaks into the fake adapter. Its real examples are the teacher's own finished images for other prompts at the same resolution; its fake examples are the student's final images. Both are lightly re-noised first, in the low-noise range where fine structure lives, and the critic reads them with an empty prompt, so it judges only whether the structure looks like the teacher's. Its push on the student is filtered to periods finer than 32 pixels β€” below that it can rebuild a face or a shelf, above it it could move layout, colour or pose, which it must not β€” and capped at a quarter of the distribution term's strength per sample. Lazy gradient regularisation and spectral normalisation keep the head from overshooting. On the largest steps, where memory is tightest, the critic and the teacher's finishing pass below take turns instead of sharing a step. ### Critics in turn One critic holds one idea of what is wrong. The artefact critic is joined by seven more on the same frozen mid-network features and under the same rules β€” lightly re-noised inputs, a push that is filtered and capped β€” each aimed at a different fault: - **A photo critic.** Half of its real examples are real photographs and half the teacher's finished images, so it learns what fine texture looks like in a photograph as well as in the teacher's rendering of one. It judges photographic prompts only, so illustration, anime and 3D renders are not pulled toward photographic grain, and its push is filtered to periods finer than 24 pixels and held lower than the artefact critic's, because photographs carry grain the teacher does not. - **A large-face critic, on photographs.** It reads only large faces, 192 pixels and up, and pushes on every step. A second face critic, reading faces of every size, trained the previous checkpoint and was taken back out (item 18 under [How I got here](#how-i-got-here)). - **A structure critic.** It judges the first call β€” the layout, before any detail β€” against the teacher's own intermediate state from the same noise, at the high noise levels where layout is decided and at periods of 32 pixels and coarser only, so a first call that blends two layouts is caught where the blend happens. - **A content critic.** It pushes on every step, weighted toward wherever the teacher put fine detail. - **A prompt critic.** Unlike the others it sees the prompt: it learns whether an image belongs to its own prompt, and is shown a mismatched prompt as a negative. - **A text critic.** It trains on the prompts that ask for lettering, and takes priority on them. - **A rollout critic.** Its real examples are the teacher's own finish from the student's second-call starting point, so the second call is judged against what the teacher would have made from the same start. All eight on every step do not fit in 24 GB, so they take turns: on each step one critic pushes, and on alternate steps the others train so none goes stale before its turn comes back. Each new head started from the artefact critic's weights and trained on its own before it was allowed to push. ### Detail terms Smaller terms sit on top, each capped relative to the distribution term so none of them can take over: - **A detail-weighted anchor.** The trajectory regression counts the fine-detail part of its error β€” everything finer than 32 pixels β€” twice, and three and a half times on the tenth of the image where the teacher has the most fine detail, so the anchor stops tolerating softness it used to average away. - **A one-sided photo floor.** On the decoded window of a photographic prompt, the student's energy at periods of 3–10 pixels may not fall below the teacher's plus the margin real photographs carry over it at those scales, judged tile by tile on the window's textured tiles. That margin is measured once from a pool of real photographs and clamped, and the term only ever pushes upward to that floor, never past it β€” so it lifts detail that is missing without adding grain that is not. - **A ceiling.** Tile by tile, detail at 3–16 pixels may not climb past 1.3Γ— the teacher's β€” the counterpart of the photo floor, and what keeps grain from building up at large sizes. - **A direction-aware term.** The spectral comparison is also made orientation by orientation, so the student's fine detail runs in the same directions as the teacher's. - **Windows on features.** On photographs, half of the decoded windows are centred on an eye, the nose or the lips of a face, so the detail terms look hardest where a face is read first. - **The second call reaching the first.** The second call's losses also flow back into the first call, at a fifth of their strength, so the first call is shaped for the finish it feeds. - **The teacher's finish as a target.** Every second step, the teacher itself runs its remaining steps starting from the student's own first-call output. The result is a finished image that shares the student's layout, and the second call is pulled gently toward it β€” a target that lines up with what the student actually drew, where the recorded trajectory might have drawn something else. - **A smoothness limit on the fake adapter.** The fake adapter's fine-detail energy is kept below the teacher's at the point where the distribution term is measured, so the difference between the two keeps pointing toward detail. ### Colour A floor holds saturation at the teacher's own level β€” on the decoded window and on the whole image, every step β€” and a ceiling 8% above it keeps it from climbing past. References that are grey or black-and-white are left alone, so a monochrome prompt is never pushed toward colour. ### A guarded running average The published weights are a running average of training (decay 0.999), which smooths out the noise of single steps. Averaging has one failure: while the training weights move between two layouts of the same prompt, their average draws both β€” a doubled subject. A guard watches a fixed set of layout probes every ten steps; a move away from the trend that comes back within six rounds is kept out of the average, and a lasting move is taken in whole β€” the guard never resets the average. Before any stretch of training goes in, the average it would make renders a fixed probe at the large sizes, and a stretch that would put a second head on the subject is held out, even when the training weights themselves drew a single one throughout. The average itself has been rebuilt twice, by hand: once after a layout change it had blended, and once from the training since the change below, after the guard had held it still through a long change of pose. ### Layout at the large sizes The doubled subjects that two steps can draw β€” a second head, two bodies joined, two poses blended β€” formed at the sizes trained most, 1280 pixels and up, and the distribution term is the one that pushes the student toward whatever the teacher would plausibly draw, which at a two-step jump can be more than one layout at once. So at every size with a 1280- or 1440-pixel side the distribution term pushes at half strength, while the anchor to the teacher's recorded trajectory keeps its full weight there: at those sizes the teacher's own layout for the same noise carries twice the share it did. The anchor also counts the coarse layout of the second call β€” everything 64 pixels and up β€” twice, since that is the call in which a second head appears. ## What the LoRA touches Rank **64**, alpha = rank, bf16, the same **228 modules** as the [4-step adapter](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA): the 8 attention and feed-forward linears of all 28 transformer blocks, plus the four global linears β€” `time_embed.linear_1`, `time_embed.linear_2`, `time_mod_proj`, `final_layer.linear` β€” that a step-count change needs most. Nothing about the base model changes. ## Training data The **13,750 recorded teacher trajectories** of the [4-step project](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA) β€” Krea 2 Turbo's own 8-step run at mu = 1.15 and guidance 0.0, every latent and velocity stored β€” serve unchanged: a 2-step chord is two of the 4-step chords end to end. 203 held-out prompts measure the student–teacher gap on unseen prompts and never receive a gradient. The spectral match, the detail terms and the critics read the teacher's finals for the training prompts, and the face critics and the windows on features use face masks computed once on those finals. The 43,044 real-photo crops of the [4-step project](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA) enter in two places: as one precomputed statistic β€” how much fine-detail energy they carry at 3–10 pixels relative to the teacher, clamped β€” which sets the photo floor, and as half of the photo critic's real examples. Only that critic's head sees them; the student and the fake adapter never do, and receive only its filtered, capped push. ## Resolutions The same 12 buckets as the [4-step adapter](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA). Since `chk00017464` the draw leans toward the larger sizes, where two steps had put the most grain β€” 1440Γ—1440 from under 2% of the samples to about 6%, 1024Γ—1024 and 768Γ—1024 to 12% each β€” and stylised prompts are drawn three times as often as their share of the pool, 12% of the samples instead of 4%: | | | | | --------- | --------- | --------- | | 512Γ—512 | 512Γ—768 | 768Γ—512 | | 768Γ—768 | 768Γ—1024 | 1024Γ—768 | | 1024Γ—1024 | 960Γ—1280 | 1280Γ—960 | | 1280Γ—1280 | 1440Γ—1280 | 1440Γ—1440 | ## Resolution sweeps The [Examples](#examples) below are all 768Γ—1024. A single resolution is not enough to judge an adapter of this kind: the shard pool it trains on is never evenly spread across buckets, and adapters carry recency bias, so one can be strong at the size it saw most while quietly softer at the ones it barely saw. Only rendering every bucket shows that, which is why every cut of this run is rendered across the buckets and why the per-resolution table under [Method](#method) above exists. [`assets/resolution_sweeps/`](assets/resolution_sweeps) holds the evidence β€” same 22 prompts (the original 15, plus seven scenes of people, animals and action added on 29 Sep 2026 to check structure), same seed, one folder per resolution, one image per prompt, so any image can be compared 1:1 with its twin in the next tree: - [`_teacher-8step/`](assets/resolution_sweeps/_teacher-8step) β€” the **official Krea 2 Turbo 8-step reference renders**: the stock model, no LoRA, at its native settings (8 steps, guidance 0.0). The teacher this LoRA is distilled from and the quality bar it is measured against. - [`_turbo-base-NO-LoRA-2step/`](assets/resolution_sweeps/_turbo-base-NO-LoRA-2step) β€” **stock Turbo at 2 steps, no LoRA**: the do-nothing baseline, what two steps look like before the adapter. Every claim on this page is checked against both trees. - [`2step-LoRA/`](assets/resolution_sweeps/2step-LoRA) β€” **this LoRA at 2 steps**, the same tree: the renders of the checkpoint published here, replaced whenever a better one ships. Which checkpoint that is today is in [`krea2_turbo_2step_rank_64_lora_checkpoint_info.md`](krea2_turbo_2step_rank_64_lora_checkpoint_info.md), and every published checkpoint's tree is kept under its own number in [`_archive/resolution_sweeps/`](_archive/resolution_sweeps). ``` assets/resolution_sweeps/ β”œβ”€β”€ _teacher-8step/ the official 8-step stock-Turbo reference renders (no LoRA) β”‚ β”œβ”€β”€ 512x512/ one folder per resolution β”‚ β”‚ β”œβ”€β”€ portrait.jpg β”‚ β”‚ β”œβ”€β”€ kingfisher.jpg β”‚ β”‚ β”œβ”€β”€ … 19 more, one per test prompt β”‚ β”‚ └── flamingos.jpg β”‚ β”œβ”€β”€ 768x1024/ β”‚ β”œβ”€β”€ 1024x1024/ β”‚ β”œβ”€β”€ 1280x1280/ β”‚ β”œβ”€β”€ 1440x1280/ …and the remaining buckets β”‚ └── 1440x1440/ β”œβ”€β”€ _turbo-base-NO-LoRA-2step/ the same tree, stock Turbo at 2 steps β€” the floor └── 2step-LoRA/ the same tree, rendered with this LoRA at 2 steps β€” the published checkpoint ``` Two ways to read them, both useful: - πŸ” **down the sweep** β€” does the adapter hold together across every resolution, or is it strong at one size and soft at others? - 🎯 **against the teacher and the floor** β€” open the same `/.jpg` under [`_teacher-8step/`](assets/resolution_sweeps/_teacher-8step) to see how close two steps with the LoRA get to the full 8-step render, and under [`_turbo-base-NO-LoRA-2step/`](assets/resolution_sweeps/_turbo-base-NO-LoRA-2step) to see what those two steps looked like without it. The resolutions are exactly the training buckets listed under [Resolutions](#resolutions). ## Hardware One **RTX 3090 (24 GB)**. The frozen base is weight-only int8; the student's checkpointed block inputs stage to pinned host memory above 0.3 megapixels; the student, the fake adapter, the spectral and detail terms, the critic and the teacher's finishing pass each build and free their own graph in turn, so their peaks never overlap; a hard memory ceiling sits below the driver's paging threshold so a step that does not fit fails loudly. A full step with every term live and every critic pushing reserves about 22.4 GB at 1440Γ—1440, of 24. The price of the objective is throughput: **about 58 training samples an hour** with every critic and detail term in the recipe, against the [4-step recipe](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA)'s 470 β€” about eight times the cost per sample, and so far a small fraction of the samples. **Where that cost comes from.** Distribution matching is simply a heavier objective than trajectory distillation. The 4-step project's recipe compared the student's own output with a teacher state that had already been recorded to disk, so a training step was one student pass plus a small adversarial head. Here every step also needs the *score* of two models at a freshly noised point: the frozen teacher's, and a second adapter's that is being trained alongside to imitate the student β€” and that second adapter takes three optimiser steps of its own per student step. The spectral and detail terms decode part of the image out of the latent to compare its texture with the teacher's, each critic reads half the network twice more, and every second step the teacher finishes the image from the student's first call. Counted in whole model runs per training sample, the difference is roughly **two there against about a dozen here**. None of that difference is the teacher generating anything: its renders were recorded once for the 4-step project and are read from disk by both. The extra work is the objective itself, and it bought the only thing that mattered. Run at two steps, the 4-step project's recipe reached a point where more training changed nothing: the measurements sat flat and every new checkpoint had the same faults as the one before β€” doubled contours on faces and limbs, soft fine texture, crowds blurred into one another. Distribution matching is the change that made each new checkpoint visibly better than the last again. ## How it is judged At regular intervals, both the live weights and their running average are pulled, merged and rendered at fixed seeds on 22 fixed prompts across four resolutions (512Γ—512, 1280Γ—1280, 1440Γ—1440, 1440Γ—1280), after a layout check at three large sizes that has to pass first; milestone checkpoints get the same render at all 12 buckets, which is where the per-resolution table above comes from. Every image is measured against the teacher's render of the same prompt and seed: distance, fine-texture energy, the 16-pixel and 8-pixel grid bands, skin and flat-region grain, saturation, faces cut out at 1:1, fixed content windows (small faces in a crowd, shop interiors seen through their windows), straight-line artefacts, a structure judge that compares the seven scenes of people, animals and action with the teacher's for missing, extra or merged parts (every flag checked by eye at 1:1), a graded judge, a pairwise preference against the teacher, and a blind rubric that scores each render on its own against the prompt's objects, counts, attributes and relations, with the teacher scored identically. Fifteen prompts drawn fresh from the prompt bank, never rendered before, are judged the same way at every checkpoint. Latent distances β€” the held-out chord gap and the two-step rollout error β€” are recorded but never used to keep or stop a run: this project's clearest lesson is that they reward blur, and a run that improved them while its pictures collapsed was stopped by its pictures. I have the last word at every gate, looking at the renders. ## Examples Every sheet below is three renders at the **same seed**: the base model as shipped, the base model at two steps _without_ the LoRA, and the same two steps _with_ it. Each panel is captioned with its own steps, CFG and NFE. Click any image for the full-size version. **NFE** = _number of function evaluations_: how many times the model itself is run, and the honest unit of cost β€” steps are not, because a step with CFG runs the model twice (once conditional, once unconditional). Turbo is CFG-free, so here NFE equals steps: the 8-step reference costs 8, and this LoRA's 2 steps cost 2. Wall-clock tracks NFE. **How to read these sheets.** Compare the **second and third panels** β€” they run the same step count and differ only by the adapter, so that pair isolates what the LoRA does. The first panel is the quality bar, not a pixel-level target: changing the step count moves the sampling trajectory by itself, so the full-step render often differs in pose and framing from both reduced-step panels regardless of whether the LoRA is loaded. Same seed throughout; the seed fixes the starting noise, not the destination. Because two steps give up more than four, each prompt also gets a **two-panel sheet against the teacher alone** β€” the LoRA's render beside the 8-step one, nothing else in the frame β€” which is the comparison this project is judged on. ### Krea 2 Turbo β€” 8 steps β†’ 2 steps One section per prompt: the three-panel comparison first, then the two-panel comparison against the teacher, then the individual renders β€” click any image for full size. ### Portrait of a young woman with freckles and windswept auburn hair, soft window light, shallow depth of field, photograph, sharp detail **3-way comparison** β€” one image, all three renders side by side [![portrait comparison](assets/thumbs/portrait_compare_turbo.jpg)](assets/portrait_compare_turbo.jpg) **Against the teacher** β€” this LoRA at 2 steps beside the 8-step render [![portrait vs teacher](assets/thumbs/portrait_compare_teacher.jpg)](assets/portrait_compare_teacher.jpg) **Individual frames** β€” click any panel to open that render full size | Turbo β€” 8 steps | Turbo β€” 2 steps, no LoRA | **Turbo β€” 2 steps + this LoRA** | | ------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------ | | [![8 steps](assets/thumbs/portrait_turbo_8step.jpg)](assets/portrait_turbo_8step.jpg) | [![2 steps](assets/thumbs/portrait_turbo_2step.jpg)](assets/portrait_turbo_2step.jpg) | [![2 steps + LoRA](assets/thumbs/portrait_turbo_2step_lora.jpg)](assets/portrait_turbo_2step_lora.jpg) | | as shipped Β· 8 NFE | the deficit this closes | strength 1.0 Β· 2 NFE | --- ### A kingfisher bird bursting out of water with spread wings, water droplets frozen mid-air, iridescent blue and orange feathers, high-speed photography **3-way comparison** β€” one image, all three renders side by side [![kingfisher comparison](assets/thumbs/kingfisher_compare_turbo.jpg)](assets/kingfisher_compare_turbo.jpg) **Against the teacher** β€” this LoRA at 2 steps beside the 8-step render [![kingfisher vs teacher](assets/thumbs/kingfisher_compare_teacher.jpg)](assets/kingfisher_compare_teacher.jpg) **Individual frames** β€” click any panel to open that render full size | Turbo β€” 8 steps | Turbo β€” 2 steps, no LoRA | **Turbo β€” 2 steps + this LoRA** | | ----------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------- | | [![8 steps](assets/thumbs/kingfisher_turbo_8step.jpg)](assets/kingfisher_turbo_8step.jpg) | [![2 steps](assets/thumbs/kingfisher_turbo_2step.jpg)](assets/kingfisher_turbo_2step.jpg) | [![2 steps + LoRA](assets/thumbs/kingfisher_turbo_2step_lora.jpg)](assets/kingfisher_turbo_2step_lora.jpg) | | as shipped Β· 8 NFE | the deficit this closes | strength 1.0 Β· 2 NFE | --- ### Rainy night city street with glowing neon shop signs and readable text, wet asphalt reflections, pedestrians with umbrellas, cinematic **3-way comparison** β€” one image, all three renders side by side [![neonstreet comparison](assets/thumbs/neonstreet_compare_turbo.jpg)](assets/neonstreet_compare_turbo.jpg) **Against the teacher** β€” this LoRA at 2 steps beside the 8-step render [![neonstreet vs teacher](assets/thumbs/neonstreet_compare_teacher.jpg)](assets/neonstreet_compare_teacher.jpg) **Individual frames** β€” click any panel to open that render full size | Turbo β€” 8 steps | Turbo β€” 2 steps, no LoRA | **Turbo β€” 2 steps + this LoRA** | | ----------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------- | | [![8 steps](assets/thumbs/neonstreet_turbo_8step.jpg)](assets/neonstreet_turbo_8step.jpg) | [![2 steps](assets/thumbs/neonstreet_turbo_2step.jpg)](assets/neonstreet_turbo_2step.jpg) | [![2 steps + LoRA](assets/thumbs/neonstreet_turbo_2step_lora.jpg)](assets/neonstreet_turbo_2step_lora.jpg) | | as shipped Β· 8 NFE | the deficit this closes | strength 1.0 Β· 2 NFE | --- ### Busy outdoor street market crowded with many people browsing colorful fruit and vegetable stalls, awnings, midday sun, wide shot, photorealistic **3-way comparison** β€” one image, all three renders side by side [![market comparison](assets/thumbs/market_compare_turbo.jpg)](assets/market_compare_turbo.jpg) **Against the teacher** β€” this LoRA at 2 steps beside the 8-step render [![market vs teacher](assets/thumbs/market_compare_teacher.jpg)](assets/market_compare_teacher.jpg) **Individual frames** β€” click any panel to open that render full size | Turbo β€” 8 steps | Turbo β€” 2 steps, no LoRA | **Turbo β€” 2 steps + this LoRA** | | --------------------------------------------------------------------------------- | --------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------- | | [![8 steps](assets/thumbs/market_turbo_8step.jpg)](assets/market_turbo_8step.jpg) | [![2 steps](assets/thumbs/market_turbo_2step.jpg)](assets/market_turbo_2step.jpg) | [![2 steps + LoRA](assets/thumbs/market_turbo_2step_lora.jpg)](assets/market_turbo_2step_lora.jpg) | | as shipped Β· 8 NFE | the deficit this closes | strength 1.0 Β· 2 NFE | --- ### Overhead shot of a rustic wood-fired pizza with bubbling melted cheese, basil leaves, charred crust, on a dark wooden table, food photography **3-way comparison** β€” one image, all three renders side by side [![pizza comparison](assets/thumbs/pizza_compare_turbo.jpg)](assets/pizza_compare_turbo.jpg) **Against the teacher** β€” this LoRA at 2 steps beside the 8-step render [![pizza vs teacher](assets/thumbs/pizza_compare_teacher.jpg)](assets/pizza_compare_teacher.jpg) **Individual frames** β€” click any panel to open that render full size | Turbo β€” 8 steps | Turbo β€” 2 steps, no LoRA | **Turbo β€” 2 steps + this LoRA** | | ------------------------------------------------------------------------------- | ------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------ | | [![8 steps](assets/thumbs/pizza_turbo_8step.jpg)](assets/pizza_turbo_8step.jpg) | [![2 steps](assets/thumbs/pizza_turbo_2step.jpg)](assets/pizza_turbo_2step.jpg) | [![2 steps + LoRA](assets/thumbs/pizza_turbo_2step_lora.jpg)](assets/pizza_turbo_2step_lora.jpg) | | as shipped Β· 8 NFE | the deficit this closes | strength 1.0 Β· 2 NFE | --- ### A young swordsman leaping through falling cherry blossoms, dynamic action pose, anime key visual, crisp linework, vivid colors **3-way comparison** β€” one image, all three renders side by side [![swordsman comparison](assets/thumbs/swordsman_compare_turbo.jpg)](assets/swordsman_compare_turbo.jpg) **Against the teacher** β€” this LoRA at 2 steps beside the 8-step render [![swordsman vs teacher](assets/thumbs/swordsman_compare_teacher.jpg)](assets/swordsman_compare_teacher.jpg) **Individual frames** β€” click any panel to open that render full size | Turbo β€” 8 steps | Turbo β€” 2 steps, no LoRA | **Turbo β€” 2 steps + this LoRA** | | --------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------- | | [![8 steps](assets/thumbs/swordsman_turbo_8step.jpg)](assets/swordsman_turbo_8step.jpg) | [![2 steps](assets/thumbs/swordsman_turbo_2step.jpg)](assets/swordsman_turbo_2step.jpg) | [![2 steps + LoRA](assets/thumbs/swordsman_turbo_2step_lora.jpg)](assets/swordsman_turbo_2step_lora.jpg) | | as shipped Β· 8 NFE | the deficit this closes | strength 1.0 Β· 2 NFE | --- ### A giant mecha standing in a rain-soaked city plaza, anime style, panel lining, glowing cockpit, dramatic low angle **3-way comparison** β€” one image, all three renders side by side [![mecha comparison](assets/thumbs/mecha_compare_turbo.jpg)](assets/mecha_compare_turbo.jpg) **Against the teacher** β€” this LoRA at 2 steps beside the 8-step render [![mecha vs teacher](assets/thumbs/mecha_compare_teacher.jpg)](assets/mecha_compare_teacher.jpg) **Individual frames** β€” click any panel to open that render full size | Turbo β€” 8 steps | Turbo β€” 2 steps, no LoRA | **Turbo β€” 2 steps + this LoRA** | | ------------------------------------------------------------------------------- | ------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------ | | [![8 steps](assets/thumbs/mecha_turbo_8step.jpg)](assets/mecha_turbo_8step.jpg) | [![2 steps](assets/thumbs/mecha_turbo_2step.jpg)](assets/mecha_turbo_2step.jpg) | [![2 steps + LoRA](assets/thumbs/mecha_turbo_2step_lora.jpg)](assets/mecha_turbo_2step_lora.jpg) | | as shipped Β· 8 NFE | the deficit this closes | strength 1.0 Β· 2 NFE | --- ### A fox in a red scarf reading a book under a mushroom, children's storybook illustration, watercolour texture, soft edges **3-way comparison** β€” one image, all three renders side by side [![storybookfox comparison](assets/thumbs/storybookfox_compare_turbo.jpg)](assets/storybookfox_compare_turbo.jpg) **Against the teacher** β€” this LoRA at 2 steps beside the 8-step render [![storybookfox vs teacher](assets/thumbs/storybookfox_compare_teacher.jpg)](assets/storybookfox_compare_teacher.jpg) **Individual frames** β€” click any panel to open that render full size | Turbo β€” 8 steps | Turbo β€” 2 steps, no LoRA | **Turbo β€” 2 steps + this LoRA** | | --------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------- | | [![8 steps](assets/thumbs/storybookfox_turbo_8step.jpg)](assets/storybookfox_turbo_8step.jpg) | [![2 steps](assets/thumbs/storybookfox_turbo_2step.jpg)](assets/storybookfox_turbo_2step.jpg) | [![2 steps + LoRA](assets/thumbs/storybookfox_turbo_2step_lora.jpg)](assets/storybookfox_turbo_2step_lora.jpg) | | as shipped Β· 8 NFE | the deficit this closes | strength 1.0 Β· 2 NFE | --- ### A curious young inventor girl with oversized goggles, 3D animated film style, subsurface skin, soft studio lighting, shallow depth of field **3-way comparison** β€” one image, all three renders side by side [![inventor comparison](assets/thumbs/inventor_compare_turbo.jpg)](assets/inventor_compare_turbo.jpg) **Against the teacher** β€” this LoRA at 2 steps beside the 8-step render [![inventor vs teacher](assets/thumbs/inventor_compare_teacher.jpg)](assets/inventor_compare_teacher.jpg) **Individual frames** β€” click any panel to open that render full size | Turbo β€” 8 steps | Turbo β€” 2 steps, no LoRA | **Turbo β€” 2 steps + this LoRA** | | ------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------ | | [![8 steps](assets/thumbs/inventor_turbo_8step.jpg)](assets/inventor_turbo_8step.jpg) | [![2 steps](assets/thumbs/inventor_turbo_2step.jpg)](assets/inventor_turbo_2step.jpg) | [![2 steps + LoRA](assets/thumbs/inventor_turbo_2step_lora.jpg)](assets/inventor_turbo_2step_lora.jpg) | | as shipped Β· 8 NFE | the deficit this closes | strength 1.0 Β· 2 NFE | --- ### A claymation chef holding a tiny cake, visible fingerprints in the clay, miniature set, tilt-shift **3-way comparison** β€” one image, all three renders side by side [![claychef comparison](assets/thumbs/claychef_compare_turbo.jpg)](assets/claychef_compare_turbo.jpg) **Against the teacher** β€” this LoRA at 2 steps beside the 8-step render [![claychef vs teacher](assets/thumbs/claychef_compare_teacher.jpg)](assets/claychef_compare_teacher.jpg) **Individual frames** β€” click any panel to open that render full size | Turbo β€” 8 steps | Turbo β€” 2 steps, no LoRA | **Turbo β€” 2 steps + this LoRA** | | ------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------ | | [![8 steps](assets/thumbs/claychef_turbo_8step.jpg)](assets/claychef_turbo_8step.jpg) | [![2 steps](assets/thumbs/claychef_turbo_2step.jpg)](assets/claychef_turbo_2step.jpg) | [![2 steps + LoRA](assets/thumbs/claychef_turbo_2step_lora.jpg)](assets/claychef_turbo_2step_lora.jpg) | | as shipped Β· 8 NFE | the deficit this closes | strength 1.0 Β· 2 NFE | --- ### A gleaming white colony ship in orbit above a turquoise ocean planet, smooth curved hull, glowing cyan engine rings, brilliant sunlight, clean sci-fi concept art, bold simple shapes, vivid colors **3-way comparison** β€” one image, all three renders side by side [![colonyship comparison](assets/thumbs/colonyship_compare_turbo.jpg)](assets/colonyship_compare_turbo.jpg) **Against the teacher** β€” this LoRA at 2 steps beside the 8-step render [![colonyship vs teacher](assets/thumbs/colonyship_compare_teacher.jpg)](assets/colonyship_compare_teacher.jpg) **Individual frames** β€” click any panel to open that render full size | Turbo β€” 8 steps | Turbo β€” 2 steps, no LoRA | **Turbo β€” 2 steps + this LoRA** | | ----------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------- | | [![8 steps](assets/thumbs/colonyship_turbo_8step.jpg)](assets/colonyship_turbo_8step.jpg) | [![2 steps](assets/thumbs/colonyship_turbo_2step.jpg)](assets/colonyship_turbo_2step.jpg) | [![2 steps + LoRA](assets/thumbs/colonyship_turbo_2step_lora.jpg)](assets/colonyship_turbo_2step_lora.jpg) | | as shipped Β· 8 NFE | the deficit this closes | strength 1.0 Β· 2 NFE | --- ### A sleek winged drone gliding between glowing futuristic skyscrapers at night, bright lit avenue far below, deep blue sky above, digital matte painting, bold clean forms, vivid colors **3-way comparison** β€” one image, all three renders side by side [![megacity comparison](assets/thumbs/megacity_compare_turbo.jpg)](assets/megacity_compare_turbo.jpg) **Against the teacher** β€” this LoRA at 2 steps beside the 8-step render [![megacity vs teacher](assets/thumbs/megacity_compare_teacher.jpg)](assets/megacity_compare_teacher.jpg) **Individual frames** β€” click any panel to open that render full size | Turbo β€” 8 steps | Turbo β€” 2 steps, no LoRA | **Turbo β€” 2 steps + this LoRA** | | ------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------ | | [![8 steps](assets/thumbs/megacity_turbo_8step.jpg)](assets/megacity_turbo_8step.jpg) | [![2 steps](assets/thumbs/megacity_turbo_2step.jpg)](assets/megacity_turbo_2step.jpg) | [![2 steps + LoRA](assets/thumbs/megacity_turbo_2step_lora.jpg)](assets/megacity_turbo_2step_lora.jpg) | | as shipped Β· 8 NFE | the deficit this closes | strength 1.0 Β· 2 NFE | --- ### A storm sorceress channelling lightning, video-game splash art, bold rim lighting, energetic brush strokes, high contrast **3-way comparison** β€” one image, all three renders side by side [![sorceress comparison](assets/thumbs/sorceress_compare_turbo.jpg)](assets/sorceress_compare_turbo.jpg) **Against the teacher** β€” this LoRA at 2 steps beside the 8-step render [![sorceress vs teacher](assets/thumbs/sorceress_compare_teacher.jpg)](assets/sorceress_compare_teacher.jpg) **Individual frames** β€” click any panel to open that render full size | Turbo β€” 8 steps | Turbo β€” 2 steps, no LoRA | **Turbo β€” 2 steps + this LoRA** | | --------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------- | | [![8 steps](assets/thumbs/sorceress_turbo_8step.jpg)](assets/sorceress_turbo_8step.jpg) | [![2 steps](assets/thumbs/sorceress_turbo_2step.jpg)](assets/sorceress_turbo_2step.jpg) | [![2 steps + LoRA](assets/thumbs/sorceress_turbo_2step_lora.jpg)](assets/sorceress_turbo_2step_lora.jpg) | | as shipped Β· 8 NFE | the deficit this closes | strength 1.0 Β· 2 NFE | --- ### A formula 1 futuristic looking racing car beefed up with a lot of technology mid-corner on a wet track, motion blur background, photorealistic motorsport photography **3-way comparison** β€” one image, all three renders side by side [![racecar comparison](assets/thumbs/racecar_compare_turbo.jpg)](assets/racecar_compare_turbo.jpg) **Against the teacher** β€” this LoRA at 2 steps beside the 8-step render [![racecar vs teacher](assets/thumbs/racecar_compare_teacher.jpg)](assets/racecar_compare_teacher.jpg) **Individual frames** β€” click any panel to open that render full size | Turbo β€” 8 steps | Turbo β€” 2 steps, no LoRA | **Turbo β€” 2 steps + this LoRA** | | ----------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------- | | [![8 steps](assets/thumbs/racecar_turbo_8step.jpg)](assets/racecar_turbo_8step.jpg) | [![2 steps](assets/thumbs/racecar_turbo_2step.jpg)](assets/racecar_turbo_2step.jpg) | [![2 steps + LoRA](assets/thumbs/racecar_turbo_2step_lora.jpg)](assets/racecar_turbo_2step_lora.jpg) | | as shipped Β· 8 NFE | the deficit this closes | strength 1.0 Β· 2 NFE | --- ### A snow leopard walking along a rocky ridge in falling snow, telephoto wildlife photograph, natural light **3-way comparison** β€” one image, all three renders side by side [![snowleopard comparison](assets/thumbs/snowleopard_compare_turbo.jpg)](assets/snowleopard_compare_turbo.jpg) **Against the teacher** β€” this LoRA at 2 steps beside the 8-step render [![snowleopard vs teacher](assets/thumbs/snowleopard_compare_teacher.jpg)](assets/snowleopard_compare_teacher.jpg) **Individual frames** β€” click any panel to open that render full size | Turbo β€” 8 steps | Turbo β€” 2 steps, no LoRA | **Turbo β€” 2 steps + this LoRA** | | ------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------ | | [![8 steps](assets/thumbs/snowleopard_turbo_8step.jpg)](assets/snowleopard_turbo_8step.jpg) | [![2 steps](assets/thumbs/snowleopard_turbo_2step.jpg)](assets/snowleopard_turbo_2step.jpg) | [![2 steps + LoRA](assets/thumbs/snowleopard_turbo_2step_lora.jpg)](assets/snowleopard_turbo_2step_lora.jpg) | | as shipped Β· 8 NFE | the deficit this closes | strength 1.0 Β· 2 NFE | --- ### A group photo of five friends standing side by side on a sunny beach, arms around each other's shoulders, all smiling at the camera; five clearly different people, each with a unique face, no twins or lookalikes: a tall bearded man in his forties, a young woman with curly red hair and freckles, an older East Asian man with grey hair and glasses, a Black woman with short natural hair, and a teenage boy with messy blond hair, photograph, sharp detail **3-way comparison** β€” one image, all three renders side by side [![group5 comparison](assets/thumbs/group5_compare_turbo.jpg)](assets/group5_compare_turbo.jpg) **Against the teacher** β€” this LoRA at 2 steps beside the 8-step render [![group5 vs teacher](assets/thumbs/group5_compare_teacher.jpg)](assets/group5_compare_teacher.jpg) **Individual frames** β€” click any panel to open that render full size | Turbo β€” 8 steps | Turbo β€” 2 steps, no LoRA | **Turbo β€” 2 steps + this LoRA** | | --------------------------------------------------------------------------------- | --------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------- | | [![8 steps](assets/thumbs/group5_turbo_8step.jpg)](assets/group5_turbo_8step.jpg) | [![2 steps](assets/thumbs/group5_turbo_2step.jpg)](assets/group5_turbo_2step.jpg) | [![2 steps + LoRA](assets/thumbs/group5_turbo_2step_lora.jpg)](assets/group5_turbo_2step_lora.jpg) | | as shipped Β· 8 NFE | the deficit this closes | strength 1.0 Β· 2 NFE | --- ### A family of four having dinner at a round wooden table, seen from directly above, plates, glasses and bowls of food, warm evening light, photograph **3-way comparison** β€” one image, all three renders side by side [![family4 comparison](assets/thumbs/family4_compare_turbo.jpg)](assets/family4_compare_turbo.jpg) **Against the teacher** β€” this LoRA at 2 steps beside the 8-step render [![family4 vs teacher](assets/thumbs/family4_compare_teacher.jpg)](assets/family4_compare_teacher.jpg) **Individual frames** β€” click any panel to open that render full size | Turbo β€” 8 steps | Turbo β€” 2 steps, no LoRA | **Turbo β€” 2 steps + this LoRA** | | ----------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------- | | [![8 steps](assets/thumbs/family4_turbo_8step.jpg)](assets/family4_turbo_8step.jpg) | [![2 steps](assets/thumbs/family4_turbo_2step.jpg)](assets/family4_turbo_2step.jpg) | [![2 steps + LoRA](assets/thumbs/family4_turbo_2step_lora.jpg)](assets/family4_turbo_2step_lora.jpg) | | as shipped Β· 8 NFE | the deficit this closes | strength 1.0 Β· 2 NFE | --- ### Three kittens sitting side by side in a wicker basket with a tall arched handle over them, all looking at the camera, soft natural light, photograph, sharp detail **3-way comparison** β€” one image, all three renders side by side [![kittens3 comparison](assets/thumbs/kittens3_compare_turbo.jpg)](assets/kittens3_compare_turbo.jpg) **Against the teacher** β€” this LoRA at 2 steps beside the 8-step render [![kittens3 vs teacher](assets/thumbs/kittens3_compare_teacher.jpg)](assets/kittens3_compare_teacher.jpg) **Individual frames** β€” click any panel to open that render full size | Turbo β€” 8 steps | Turbo β€” 2 steps, no LoRA | **Turbo β€” 2 steps + this LoRA** | | ------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------ | | [![8 steps](assets/thumbs/kittens3_turbo_8step.jpg)](assets/kittens3_turbo_8step.jpg) | [![2 steps](assets/thumbs/kittens3_turbo_2step.jpg)](assets/kittens3_turbo_2step.jpg) | [![2 steps + LoRA](assets/thumbs/kittens3_turbo_2step_lora.jpg)](assets/kittens3_turbo_2step_lora.jpg) | | as shipped Β· 8 NFE | the deficit this closes | strength 1.0 Β· 2 NFE | --- ### Two golden retrievers playing tug of war with a red rope on a green lawn, action shot, photograph, sharp detail **3-way comparison** β€” one image, all three renders side by side [![dogs2 comparison](assets/thumbs/dogs2_compare_turbo.jpg)](assets/dogs2_compare_turbo.jpg) **Against the teacher** β€” this LoRA at 2 steps beside the 8-step render [![dogs2 vs teacher](assets/thumbs/dogs2_compare_teacher.jpg)](assets/dogs2_compare_teacher.jpg) **Individual frames** β€” click any panel to open that render full size | Turbo β€” 8 steps | Turbo β€” 2 steps, no LoRA | **Turbo β€” 2 steps + this LoRA** | | ------------------------------------------------------------------------------- | ------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------ | | [![8 steps](assets/thumbs/dogs2_turbo_8step.jpg)](assets/dogs2_turbo_8step.jpg) | [![2 steps](assets/thumbs/dogs2_turbo_2step.jpg)](assets/dogs2_turbo_2step.jpg) | [![2 steps + LoRA](assets/thumbs/dogs2_turbo_2step_lora.jpg)](assets/dogs2_turbo_2step_lora.jpg) | | as shipped Β· 8 NFE | the deficit this closes | strength 1.0 Β· 2 NFE | --- ### A horse and rider jumping over a wooden fence at a show jumping event, side view, sports photography, sharp detail **3-way comparison** β€” one image, all three renders side by side [![horserider comparison](assets/thumbs/horserider_compare_turbo.jpg)](assets/horserider_compare_turbo.jpg) **Against the teacher** β€” this LoRA at 2 steps beside the 8-step render [![horserider vs teacher](assets/thumbs/horserider_compare_teacher.jpg)](assets/horserider_compare_teacher.jpg) **Individual frames** β€” click any panel to open that render full size | Turbo β€” 8 steps | Turbo β€” 2 steps, no LoRA | **Turbo β€” 2 steps + this LoRA** | | ----------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------- | | [![8 steps](assets/thumbs/horserider_turbo_8step.jpg)](assets/horserider_turbo_8step.jpg) | [![2 steps](assets/thumbs/horserider_turbo_2step.jpg)](assets/horserider_turbo_2step.jpg) | [![2 steps + LoRA](assets/thumbs/horserider_turbo_2step_lora.jpg)](assets/horserider_turbo_2step_lora.jpg) | | as shipped Β· 8 NFE | the deficit this closes | strength 1.0 Β· 2 NFE | --- ### Close-up of a pianist's two hands playing the keys of a grand piano, dramatic side light, photograph, sharp detail **3-way comparison** β€” one image, all three renders side by side [![pianist comparison](assets/thumbs/pianist_compare_turbo.jpg)](assets/pianist_compare_turbo.jpg) **Against the teacher** β€” this LoRA at 2 steps beside the 8-step render [![pianist vs teacher](assets/thumbs/pianist_compare_teacher.jpg)](assets/pianist_compare_teacher.jpg) **Individual frames** β€” click any panel to open that render full size | Turbo β€” 8 steps | Turbo β€” 2 steps, no LoRA | **Turbo β€” 2 steps + this LoRA** | | ----------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------- | | [![8 steps](assets/thumbs/pianist_turbo_8step.jpg)](assets/pianist_turbo_8step.jpg) | [![2 steps](assets/thumbs/pianist_turbo_2step.jpg)](assets/pianist_turbo_2step.jpg) | [![2 steps + LoRA](assets/thumbs/pianist_turbo_2step_lora.jpg)](assets/pianist_turbo_2step_lora.jpg) | | as shipped Β· 8 NFE | the deficit this closes | strength 1.0 Β· 2 NFE | --- ### A flock of pink flamingos standing in shallow turquoise water with their reflections, wildlife photography, sharp detail **3-way comparison** β€” one image, all three renders side by side [![flamingos comparison](assets/thumbs/flamingos_compare_turbo.jpg)](assets/flamingos_compare_turbo.jpg) **Against the teacher** β€” this LoRA at 2 steps beside the 8-step render [![flamingos vs teacher](assets/thumbs/flamingos_compare_teacher.jpg)](assets/flamingos_compare_teacher.jpg) **Individual frames** β€” click any panel to open that render full size | Turbo β€” 8 steps | Turbo β€” 2 steps, no LoRA | **Turbo β€” 2 steps + this LoRA** | | --------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------- | | [![8 steps](assets/thumbs/flamingos_turbo_8step.jpg)](assets/flamingos_turbo_8step.jpg) | [![2 steps](assets/thumbs/flamingos_turbo_2step.jpg)](assets/flamingos_turbo_2step.jpg) | [![2 steps + LoRA](assets/thumbs/flamingos_turbo_2step_lora.jpg)](assets/flamingos_turbo_2step_lora.jpg) | | as shipped Β· 8 NFE | the deficit this closes | strength 1.0 Β· 2 NFE | --- ## Notes and limitations - 🎯 **Krea 2 Turbo only**, at **2 steps**, guidance **0.0** (cfg 1.0 in ComfyUI), **mu = 1.15** β€” the two training sigmas are anchored to that grid. - πŸ§ͺ **Not a finished adapter.** Usable at 2 steps for previews and drafts; not the [4-step LoRA](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA)'s quality, which remains the recommendation for quality renders. Training continues, and a later checkpoint replaces this file only when the sweeps and I visually agree it is better. ## What's next Training continues from this checkpoint, one recipe change at a time, each kept only if the pictures do not degrade at any resolution β€” aiming at the best quality two steps can give, not at matching the [4-step LoRA](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA). Next, aimed at what this checkpoint still gets wrong: small subjects in wide scenes; the finest edges and the texture of skin at the largest sizes, lifted to the teacher's level without bringing the grain back; the layout at the largest sizes, where two plausible poses of the same subject can meet; the fine detail of close-up faces β€” eyes and skin texture; and the tactile surface of stylised materials such as clay β€” each held to the rule that none of it may cost the structure, the colour and the clean large sizes this checkpoint gained. A better checkpoint replaces this one when the sweeps and I visually agree, the same discipline as the [4-step adapter](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA); until then the [4-step adapter](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA) remains the recommendation for quality renders, and this one is the fast preview. ## License The adapter is a derivative of [Krea 2 Turbo](https://huggingface.co/krea/Krea-2-Turbo) and is covered by the **Krea 2 Community License Agreement** (`LICENSE.pdf` in this repository), as the [4-step adapter](https://huggingface.co/lvladikov/Krea2-Turbo-Distill-4step-LoRA) is. The attribution notice the license requires of a derivative ships with the files.