The model was briefly trained (50k steps and 5k datasets), focusing mainly on beautiful female faces. As a result, the anatomical details (hands and legs) are not very accurate. However, the facial quality has reached a quite pleasing level.So please don’t expect perfection. Stay tuned for V2, in which I will further train it to improve anatomy.I used an L4 GPU on Colab to generate images with ComfyUI via ComfyUI_ExtraModels, using a workflow directly from Nvidia. Once the model is fully loaded, it takes only 7–8 seconds per image (L4 is slightly faster than RTX 3060 12GB).model dir = ComfyUI/models/checkpointsRecommended settings: Step: 25–40 (recommended: 30) Sampler: euler + sampleEnjoy, everyone!

Files with SHA256: c58c5f1e6012a0bc923853609a5ba1977ebc66d5de67b3ef0e7ad994d21e6e29