InfiniteYou
Local AI image generation uses FLUX and a reference face to create personalized photos. Supports ComfyUI; model weights are restricted to academic research.
InfiniteYou generates images from a reference photo and a text prompt while preserving the person's facial identity. It's a local AI framework for researchers and creators exploring personalized portraits with FLUX, with an official ComfyUI integration for visual workflows.
Its InfuseNet component adds identity information to the image model while retaining its ability to follow prompts. The training approach targets recognizable faces without a pasted-on appearance. Two model variants offer different priorities: aes_stage2 favors prompt alignment and aesthetics, while sim_stage1 favors closer facial resemblance.
InfiniteYou works with FLUX.1-dev and supports replacement base models such as FLUX.1-schnell. ControlNets and LoRAs provide additional control and customization. IP-Adapter can apply a reference image's style, while OminiControl supports images that combine a personalized identity with a personalized object.
Local inference uses a CUDA GPU. The full model needs around 43GB of peak VRAM; CPU offloading and 8-bit quantization together reduce that to around 16GB with similar reported performance. A hosted Hugging Face demo provides a separate way to try image generation online.
The code is open source under Apache 2.0. The model weights use CC BY-NC 4.0 and are restricted to academic research. FLUX, InsightFace face models and any added LoRAs carry their own licenses.
