InfiniteYou | Identity-Preserving Face Generation

ComfyUI InfiniteYou Description

1. What is the InfiniteYou Workflow?

The ComfyUI InfiniteYou workflow integrates ByteDance's advanced identity-preserving model into the environment. Built on the FLUX diffusion transformer and powered by InfuseNet, this technology enables flexible and high-fidelity image generation while preserving identity features. The InfiniteYou toolkit includes two tailored workflows: Face Combine and Zero-Shot Task, each designed for different creative goals within the InfiniteYou system.

2. Benefits of InfiniteYou:

• Identity Preservation: InfiniteYou retains facial features even in stylistically transformed prompts.
• Aesthetic Quality: The aes_stage2 mode offers enhanced prompt-image alignment and beauty.
• Workflow Variety: InfiniteYou includes both Face Combine and Zero-Shot Task for different use cases.
• Parameter Control: InfiniteYou allows adjustments to guidance, fusion weights, and control timing for precise generation.
• Plug-and-Play Integration: Seamlessly integrates with standard ComfyUI workflows.

3. How to Use the InfiniteYou Workflow

IMPORTANT NOTE: When you first load the InfiniteYou workflow, you will see both Zero-Shot Task and Face Combine sections. These are two separate workflows that operate independently. While technically both can run at the same time, they are designed to be used as separate tools for different purposes, not as connected steps in a single process.

3.1 Generation Methods with InfiniteYou

Example Setup for InfiniteYou:

1. Prepare inputs: In Load Image nodes:
   • upload two reference faces for Face Combine
   • or upload one reference face image for Zero-Shot Task In CLIP Text Encode nodes:
   • describe desired scene (e.g., "a boy, 10 years old, handsome in the classroom")
   • negative prompt is optional
2. Click Queue Prompt button to run the InfiniteYou workflow
3. In Save Image: get your output

Face Combine Workflow (Blending Two Faces)

• Best for: Combining facial features from two identities with InfiniteYou's strong identity control
• Characteristics:
  • Merges identity from two images
  • Controlled blending with weights
  • Precise start and end fusion timing

Zero-Shot Task Workflow (Single Image + Prompt)

• Best for: Generating portraits from a single identity and a rich text prompt
• Characteristics:
  • High identity fidelity with sim_stage1
  • No need for dual-face comparison
  • Text-guided face recrafting

3.2 Parameter Reference for InfiniteYou

Face Combine Node: This node blends facial features from two images.

• adapter_file: Specifies the model file used for identity blending (e.g., aes_stage2_img_proj.bin).
• weight: Controls how strongly the fusion blends both faces.
• balance: Adjusts which image contributes more to the final face.
• start_at: When the fusion starts in the generation timeline.
• end_at: When the fusion ends during generation.
• fixed_face_pose: Locks face pose if true, allows variation if false.

Apply Node: Applies the InfiniteYou model to a single reference image.

• adapter_file: Defines which stage model is used.
• weight: Intensity of identity preservation.
• start_at: Start of effect application during generation.
• end_at: End of effect application.
• fixed_face_pose: If true, keeps original pose rigid.

FluxGuidance / BasicGuider: Applies extra influence to preserve identity or control prompt alignment with InfiniteYou.

• guidance: Strength of conditioning—higher = more control, lower = more variety in outputs.

Samplers: Used to control how the image is created from noise in InfiniteYou.

• sampler_name: Algorithm to generate the image (e.g., euler).
• steps: Number of iterations to refine the image.
• denoise: How much noise is removed: higher = cleaner image.

3.3. Advanced Optimization with InfiniteYou

Switching Models:

• aes_stage2: Better text-image coherence and style (after fine-tuning).
• sim_stage1: More accurate face identity retention (pre-fine-tuning).
• Always update both adapter_file and control_net file together when switching InfiniteYou model modes.

Prompt Tips for InfiniteYou:

• Add specific identity cues like "a woman", "an elderly man", etc. to improve output alignment
• Be clear and concise with subject and setting for optimal results

More Information about InfiniteYou

For additional details and development references:

• InfiniteYou original model by ByteDance
• Implementation by ZenAI-Vietnam

Acknowledgements

This workflow is powered by InfiniteYou, developed by ByteDance Intelligent Creation. The integration is provided by ZenAI-Vietnam, including tailored workflows and model conversions that enable zero-shot and multi-reference identity-preserving generation. Full credit goes to the original authors for their work.