Instructions to use hodgy/shantae-krea2-turbo-v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use hodgy/shantae-krea2-turbo-v1 with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("krea/Krea-2-Turbo", dtype=torch.bfloat16, device_map="cuda") pipe.load_lora_weights("hodgy/shantae-krea2-turbo-v1") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Inference
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- Draw Things
- DiffusionBee
Upload final Shantae Krea 2 Turbo LoRA and samples
Browse files- .gitattributes +2 -0
- README.md +61 -0
- config.yaml +126 -0
- samples/sample_step3000_01.jpg +0 -0
- samples/sample_step3000_02.jpg +3 -0
- samples/sample_step3000_03.jpg +0 -0
- samples/sample_step3000_04.jpg +0 -0
- samples/sample_step3000_05.jpg +0 -0
- samples/sample_step3000_06.jpg +0 -0
- samples/sample_step3000_07.jpg +0 -0
- samples/sample_step3000_08.jpg +0 -0
- samples/sample_step3000_09.jpg +3 -0
- samples/sample_step3000_10.jpg +0 -0
- shantae_krea2_turbo_v1_step3000.safetensors +3 -0
.gitattributes
CHANGED
|
@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
| 33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
| 33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
| 36 |
+
samples/sample_step3000_02.jpg filter=lfs diff=lfs merge=lfs -text
|
| 37 |
+
samples/sample_step3000_09.jpg filter=lfs diff=lfs merge=lfs -text
|
README.md
ADDED
|
@@ -0,0 +1,61 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: diffusers
|
| 3 |
+
pipeline_tag: text-to-image
|
| 4 |
+
base_model: krea/Krea-2-Turbo
|
| 5 |
+
tags:
|
| 6 |
+
- lora
|
| 7 |
+
- krea-2-turbo
|
| 8 |
+
- ai-toolkit
|
| 9 |
+
- text-to-image
|
| 10 |
+
- portrait
|
| 11 |
+
---
|
| 12 |
+
|
| 13 |
+
# Shantae Krea 2 Turbo LoRA v1
|
| 14 |
+
|
| 15 |
+
Identity LoRA trained for **Krea 2 Turbo** with Ostris AI Toolkit.
|
| 16 |
+
|
| 17 |
+
## Usage
|
| 18 |
+
|
| 19 |
+
- **Trigger word:** `shantae`
|
| 20 |
+
- **Base model:** `krea/Krea-2-Turbo`
|
| 21 |
+
- **LoRA file:** [`shantae_krea2_turbo_v1_step3000.safetensors`](./shantae_krea2_turbo_v1_step3000.safetensors)
|
| 22 |
+
- **Recommended starting point:** 9 steps, guidance 1, LoRA strength 0.8–1.0
|
| 23 |
+
|
| 24 |
+
Example prompt:
|
| 25 |
+
|
| 26 |
+
```text
|
| 27 |
+
shantae, close-up portrait, natural window light, detailed skin, shallow depth of field
|
| 28 |
+
```
|
| 29 |
+
|
| 30 |
+
## Sample images
|
| 31 |
+
|
| 32 |
+
All samples below were generated from the final step-3000 checkpoint at 1024×1024.
|
| 33 |
+
|
| 34 |
+
| | |
|
| 35 |
+
|---|---|
|
| 36 |
+
|  |  |
|
| 37 |
+
|  |  |
|
| 38 |
+
|  |  |
|
| 39 |
+
|  |  |
|
| 40 |
+
|  |  |
|
| 41 |
+
|
| 42 |
+
## Training details
|
| 43 |
+
|
| 44 |
+
| Setting | Value |
|
| 45 |
+
|---|---|
|
| 46 |
+
| Training steps | 3,000 |
|
| 47 |
+
| LoRA rank / alpha | 32 / 32 |
|
| 48 |
+
| Precision | BF16 |
|
| 49 |
+
| Optimizer | AdamW8Bit |
|
| 50 |
+
| Learning rate | 0.0001 |
|
| 51 |
+
| Batch size | 1 |
|
| 52 |
+
| Gradient accumulation | 1 |
|
| 53 |
+
| Resolution buckets | 512, 768, 1024 |
|
| 54 |
+
| Scheduler | FlowMatch |
|
| 55 |
+
| Sample steps / guidance | 9 / 1 |
|
| 56 |
+
|
| 57 |
+
The original AI Toolkit configuration is included as [`config.yaml`](./config.yaml).
|
| 58 |
+
|
| 59 |
+
## Notes
|
| 60 |
+
|
| 61 |
+
Generated images may reproduce limitations or biases present in the training data and base model. Review outputs before use, especially in public or commercial contexts.
|
config.yaml
ADDED
|
@@ -0,0 +1,126 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
job: extension
|
| 2 |
+
meta:
|
| 3 |
+
name: shantae_krea2_turbo_v1
|
| 4 |
+
version: '1.0'
|
| 5 |
+
config:
|
| 6 |
+
name: shantae_krea2_turbo_v1
|
| 7 |
+
process:
|
| 8 |
+
- save:
|
| 9 |
+
dtype: bf16
|
| 10 |
+
save_every: 250
|
| 11 |
+
push_to_hub: false
|
| 12 |
+
save_format: diffusers
|
| 13 |
+
max_step_saves_to_keep: 4
|
| 14 |
+
type: diffusion_trainer
|
| 15 |
+
model:
|
| 16 |
+
arch: krea2:turbo
|
| 17 |
+
qtype: qfloat8
|
| 18 |
+
compile: false
|
| 19 |
+
low_vram: true
|
| 20 |
+
qtype_te: qfloat8
|
| 21 |
+
quantize: true
|
| 22 |
+
quantize_te: true
|
| 23 |
+
model_kwargs: {}
|
| 24 |
+
name_or_path: krea/Krea-2-Turbo
|
| 25 |
+
layer_offloading: false
|
| 26 |
+
assistant_lora_path: ostris/krea2_turbo_training_adapter/krea2_turbo_training_adapter_v1.safetensors
|
| 27 |
+
layer_offloading_transformer_percent: 1
|
| 28 |
+
layer_offloading_text_encoder_percent: 1
|
| 29 |
+
train:
|
| 30 |
+
lr: 0.0001
|
| 31 |
+
dtype: bf16
|
| 32 |
+
steps: 3000
|
| 33 |
+
loss_type: mse
|
| 34 |
+
optimizer: adamw8bit
|
| 35 |
+
batch_size: 1
|
| 36 |
+
ema_config:
|
| 37 |
+
use_ema: false
|
| 38 |
+
ema_decay: 0.99
|
| 39 |
+
train_unet: true
|
| 40 |
+
timestep_type: linear
|
| 41 |
+
noise_scheduler: flowmatch
|
| 42 |
+
content_or_style: balanced
|
| 43 |
+
disable_sampling: false
|
| 44 |
+
optimizer_params:
|
| 45 |
+
weight_decay: 0.0001
|
| 46 |
+
skip_first_sample: false
|
| 47 |
+
force_first_sample: false
|
| 48 |
+
train_text_encoder: false
|
| 49 |
+
unload_text_encoder: false
|
| 50 |
+
cache_text_embeddings: false
|
| 51 |
+
gradient_accumulation: 1
|
| 52 |
+
switch_boundary_every: 1
|
| 53 |
+
gradient_checkpointing: true
|
| 54 |
+
diff_output_preservation: false
|
| 55 |
+
bypass_guidance_embedding: false
|
| 56 |
+
diff_output_preservation_class: person
|
| 57 |
+
diff_output_preservation_multiplier: 1
|
| 58 |
+
device: cuda
|
| 59 |
+
sample:
|
| 60 |
+
fps: 1
|
| 61 |
+
neg: ''
|
| 62 |
+
seed: 42
|
| 63 |
+
width: 1024
|
| 64 |
+
height: 1024
|
| 65 |
+
sampler: flowmatch
|
| 66 |
+
samples:
|
| 67 |
+
- prompt: shantae, close-up portrait, natural window light, detailed skin, shallow
|
| 68 |
+
depth of field
|
| 69 |
+
- prompt: shantae, waist-up portrait outdoors at golden hour, casual outfit,
|
| 70 |
+
cinematic photography
|
| 71 |
+
- prompt: shantae, full-body fashion portrait on a city street, confident pose,
|
| 72 |
+
soft daylight
|
| 73 |
+
- prompt: shantae, seated in a modern cafe, candid expression, warm ambient
|
| 74 |
+
lighting
|
| 75 |
+
- prompt: shantae, studio portrait against a clean neutral backdrop, professional
|
| 76 |
+
softbox lighting
|
| 77 |
+
- prompt: shantae, walking along a beach at sunset, wind in hair, natural candid
|
| 78 |
+
photograph
|
| 79 |
+
- prompt: shantae, elegant evening outfit at a rooftop venue, city lights bokeh,
|
| 80 |
+
editorial photo
|
| 81 |
+
- prompt: shantae, cozy sweater in a bright apartment, relaxed pose, lifestyle
|
| 82 |
+
photography
|
| 83 |
+
- prompt: shantae, athletic outfit in a park, dynamic pose, crisp morning light
|
| 84 |
+
- prompt: shantae, black leather jacket under neon city lights, cinematic night
|
| 85 |
+
portrait
|
| 86 |
+
walk_seed: true
|
| 87 |
+
num_frames: 1
|
| 88 |
+
sample_every: 500
|
| 89 |
+
sample_steps: 9
|
| 90 |
+
guidance_scale: 1
|
| 91 |
+
sample_start_step: 0
|
| 92 |
+
logging:
|
| 93 |
+
log_every: 1
|
| 94 |
+
use_ui_logger: true
|
| 95 |
+
network:
|
| 96 |
+
type: lora
|
| 97 |
+
linear: 32
|
| 98 |
+
lokr_factor: -1
|
| 99 |
+
linear_alpha: 32
|
| 100 |
+
lokr_full_rank: true
|
| 101 |
+
network_kwargs:
|
| 102 |
+
ignore_if_contains: []
|
| 103 |
+
datasets:
|
| 104 |
+
- flip_x: false
|
| 105 |
+
flip_y: false
|
| 106 |
+
is_reg: false
|
| 107 |
+
controls: []
|
| 108 |
+
mask_path: null
|
| 109 |
+
num_frames: 1
|
| 110 |
+
resolution:
|
| 111 |
+
- 512
|
| 112 |
+
- 768
|
| 113 |
+
- 1024
|
| 114 |
+
caption_ext: txt
|
| 115 |
+
folder_path: /app/ai-toolkit/datasets/shantae_v1
|
| 116 |
+
num_repeats: 1
|
| 117 |
+
mask_min_value: 0.1
|
| 118 |
+
network_weight: 1
|
| 119 |
+
default_caption: shantae
|
| 120 |
+
caption_dropout_rate: 0.05
|
| 121 |
+
cache_latents_to_disk: true
|
| 122 |
+
shrink_video_to_frames: true
|
| 123 |
+
trigger_word: shantae
|
| 124 |
+
sqlite_db_path: /app/ai-toolkit/aitk_db.db
|
| 125 |
+
training_folder: /app/ai-toolkit/output
|
| 126 |
+
performance_log_every: 10
|
samples/sample_step3000_01.jpg
ADDED
|
samples/sample_step3000_02.jpg
ADDED
|
Git LFS Details
|
samples/sample_step3000_03.jpg
ADDED
|
samples/sample_step3000_04.jpg
ADDED
|
samples/sample_step3000_05.jpg
ADDED
|
samples/sample_step3000_06.jpg
ADDED
|
samples/sample_step3000_07.jpg
ADDED
|
samples/sample_step3000_08.jpg
ADDED
|
samples/sample_step3000_09.jpg
ADDED
|
Git LFS Details
|
samples/sample_step3000_10.jpg
ADDED
|
shantae_krea2_turbo_v1_step3000.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:12cbcd34a5270afefb178053d038b22a1cb3786f1a7b9b69b9da629248058dff
|
| 3 |
+
size 228587808
|