| --- |
| license: creativeml-openrail-m |
| tags: [stable-diffusion, sd1.5, lcm, mnn, tokforge, image-generation, cinematic, photoreal, epicrealism] |
| --- |
| |
| ## TokForge |
|
|
| - **Website:** https://tokforge.ai |
| - **Discord:** https://discord.gg/Acv3CBtfVm |
| - **Google Play:** https://play.google.com/store/apps/details?id=dev.tokforge |
| - **iOS TestFlight:** https://testflight.apple.com/join/jnufjzRr |
|
|
| Runs on-device in the TokForge app. |
|
|
| # TokForge SD1.5-LCM — Cinematic (epiCRealism) |
|
|
| An **epiCRealism + LCM** SD1.5 MNN bundle — the TokForge `SD15_LCM_MNN` fast few-step image model, |
| but with the candid/film-look photoreal **epiCRealism** checkpoint as the base instead of |
| DreamShaper-7. Renders a coherent, cinematic-photoreal 512px image in 4 steps on the in-process MNN |
| diffusion engine — full MNN speed, the film look baked into the UNet weights. |
|
|
| This is the **Cinematic** tier of the TokForge styled image catalog — the twin photoreal style to |
| the Photoreal (Realistic Vision V5.1) tier. epiCRealism leans toward a candid, cinematic film look: |
| naturalistic lighting and colour, true-to-life skin and texture, lifelike people and scenes. It won |
| the candid/film-look people category in TokForge's internal photoreal-base A/B. Verified on-device |
| (OnePlus D9500 CPU, 4 steps) rendering a clean candid/film-look portrait at MNN speed. |
|
|
| 9-file MNN bundle: `unet.mnn(.weight)` (INT8, epiCRealism + LCM fused), reference SD1.5 |
| f16 `text_encoder.mnn(.weight)` + `vae_decoder.mnn(.weight)`, `vocab.json`, `merges.txt`, |
| `alphas.txt`. |
|
|
| ## Provenance & licenses |
| - **Base model:** [emilianJR/epiCRealism](https://huggingface.co/emilianJR/epiCRealism) |
| — **CreativeML-OpenRAIL-M** (commercial-OK; redistribution of derivatives is licensed under |
| OpenRAIL-M §III.4). epiCRealism is a photoreal/film-look SD1.5 checkpoint. |
| - **LCM adapter:** [latent-consistency/lcm-lora-sdv1-5](https://huggingface.co/latent-consistency/lcm-lora-sdv1-5) |
| — openrail++ (UNet-only consistency adapter, fused to keep the 4-step floor). |
| - **CLIP / VAE / tokenizer:** the standard SD1.5 reference assets (CLIP ViT-L/14 text encoder + |
| SD1.5 VAE) reused verbatim from the TokForge base LCM bundle. This matches the other TokForge |
| SD1.5-LCM bundles' 9-file layout and dodges the libDIF CLIP-trace issue with freshly-exported CLIP. |
|
|
| ## Modifications (OpenRAIL-M §III "mark modified") |
| This bundle is a **modified derivative** of epiCRealism: |
| 1. The LCM consistency adapter (`lcm-lora-sdv1-5`) is fused into the UNet (fp32 ΔW via |
| `diffusers.fuse_lora()`) so the model is coherent at 4-8 steps with CFG≈1.0. |
| 2. The fused UNet is exported to ONNX and converted to an **INT8-quantized MNN** model |
| (`MNNConvert`, asymmetric 8-bit weight quant) for on-device CPU inference. |
| 3. CLIP and VAE are the SD1.5 reference MNN assets, not epiCRealism's own. |
|
|
| No retraining or fine-tuning of the original weights was performed beyond the LCM fuse + quant. |
|
|
| ## Use restrictions (OpenRAIL-M Attachment A) |
| Use of this model is subject to the CreativeML Open RAIL-M license, including the |
| **Attachment A use-based restrictions**: you agree not to use the model, or any derivative, to |
| violate any law; to exploit/harm minors; to generate or disseminate verifiably false information |
| to harm others; to generate or disseminate personal identifiable information to harm someone; to |
| defame, disparage or harass others; for fully automated decision-making that adversely affects |
| legal rights or creates binding obligations; for discrimination or harm to individuals or groups |
| based on protected characteristics; to exploit vulnerabilities of a specific group; to generate |
| non-consensual or false content about individuals; or to provide medical advice/interpretation of |
| medical results as a substitute for professional advice. See the full license text: |
| https://huggingface.co/spaces/CompVis/stable-diffusion-license |
|
|
| These use-based restrictions **propagate** to all who use or redistribute this bundle. |
|
|
| ## Attribution |
| - epiCRealism — © emilianJR, CreativeML-OpenRAIL-M. |
| - Latent Consistency Model LoRA (SD1.5) — Latent Consistency team, openrail++. |
| - Packaged for on-device MNN inference by TokForge (dev.tokforge). |
|
|
