ComfyUI InfiniteYou
https://github.com/bytedance/comfyui_infiniteyou- Category
- AI Tools
- Rank
- No. 944Tools index
- Pricing
- Open Source
- Type
- TOOL
- Builder
- bytedance
- GitHub
- 299 stars
- Date
About
Official ComfyUI node for InfiniteYou with FLUX — identity-preserving image generation and personalization (ICCV 2025 Highlight).
What it does
It turns a reference face into generation conditioning for ComfyUI. The pipeline detects the largest face, extracts its landmarks and identity embedding, projects that embedding into FLUX conditioning, then injects it through a control network. A pose image, region mask, strength, and active sampling interval can further guide the result.
Why it's ranked here
This is a strong specialist choice because it connects the complete identity-conditioning pipeline to native ComfyUI workflows. Official registry distribution, automatic downloads for core identity models, FP8 support, and supplied single-person and two-person examples reduce setup work. Its hardware and licensing demands sharply limit broader use.
What's good
The controls expose useful production choices instead of hiding the pipeline. Users can select face-analysis resolution and provider, choose eight or sixteen projection tokens, adjust conditioning strength and timing, and apply a region mask. Face detection retries several resolutions automatically, while missing identity and control models can download at runtime.
Tradeoffs
Hardware is the clearest barrier: documented peak VRAM is about 43 GB for BF16 and 24 GB for FP8. FP8 still requires manually placed FLUX, VAE, and text-encoder models. The code is Apache-licensed, but the released model is noncommercial and designated for academic research. The two-person workflow is explicitly reference-only.
How to use it well
Use it for research workflows already centered on ComfyUI and FLUX, especially controlled portrait recrafting from a clear reference face. Start from the supplied single-person workflow, add pose guidance when composition matters, and reserve masked blending for two-person experiments. It does not replace ComfyUI, supply every FLUX component, or provide commercial model rights.
Technical notes+
__init__.py exports the ComfyUI node mappings from nodes.py. In nodes.py, IDEmbeddingModelLoader initializes InsightFace AntelopeV2 and ArcFace, constructs the Resampler from resampler.py, and downloads missing assets through huggingface_hub. ExtractIDEmbedding selects the largest detected face and produces conditioning, while InfuseNetApply attaches identity conditioning, pose input, optional masks, strength, and sampling bounds. infuse_net.py implements InfuseNet as a ComfyUI ControlNet and injects identity embeddings into FLUX control conditioning. utils.py normalizes the aligned face and calls .cuda() directly, so the exposed CPU face-analysis provider does not make the complete embedding path CPU-only. examples/infinite_you_workflow.json and examples/multi_id_infinite_you_workflow.json provide executable graph examples. pyproject.toml defines registry metadata and dependencies, while .github/workflows/publish.yml publishes metadata changes to the Comfy registry.
Observed
- Code license
- Apache License 2.0
- Model license
- Creative Commons Attribution-NonCommercial 4.0, designated for academic research
- Primary language
- Python
- Interface
- Native ComfyUI custom nodes and importable ComfyUI workflow graphs
- Installation surface
- Official Comfy Registry and ComfyUI Node Manager, or repository clone with pip dependencies
- Core dependencies
- facexlib, ONNX Runtime, InsightFace, OpenCV, and Hugging Face Hub
- Model acquisition
- Core InfiniteYou and AntelopeV2 assets can download automatically at runtime
- Precision options
- BF16 and FP8 InfiniteYou model variants are supported
Read from README.md, pyproject.toml, requirements.txt, __init__.py, nodes.py, utils.py, resampler.py, infuse_net.py, LICENSE, examples/infinite_you_workflow.json, examples/multi_id_infinite_you_workflow.json, .github/workflows/publish.yml.
What it can do
Generate identity-preserving images
Reference images of a person and text prompts → New images maintaining the person's identity
Create personalized image variations
Source image and style/context descriptions → Customized images with preserved facial features
Process images through FLUX model
Image data and generation parameters → High-quality generated images
Integrate with ComfyUI workflows
ComfyUI node connections and parameters → Processed images within ComfyUI pipeline
Preserve facial identity across different scenes
Person's reference photos and scene descriptions → Images showing the same person in new contexts
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.