UNO ByteDance | Consistent Subject & Object Generation

ComfyUI UNO Description

1. What is ByteDance's ComfyUI UNO?

The ComfyUI UNO workflow integrates ByteDance's Universal aNd cOntrollable (UNO) model into the ComfyUI environment. Developed by ByteDance's Intelligent Creation Team, UNO represents a significant advancement in subject-driven image generation technology based on the innovative "Less-to-More" generalization paradigm. UNO enables flexible and high-fidelity image generation while preserving identity features of reference subjects with remarkable consistency across various contexts and scenarios.

2. Benefits of ByteDance's ComfyUI UNO:

• Superior Subject Consistency: UNO maintains exceptional subject similarity across different context.
• Subject-Object Control: Unlike many alternatives, UNO excels at handling both subject references and object references simultaneously without identity confusion.
• Universal Compatibility: UNO works with diverse types of subjects - people, objects, toys, logos, and more.
• Resource Efficiency: UNO delivers professional-quality results even with limited VRAM resources (works well with 24GB).
• Flexible Implementation: UNO includes workflows for both single-subject and subject-object generation.

3. How to Use ByteDance's ComfyUI UNO

3.1 Workflow Setup

The UNO workflow has two main configurations:

Top Group - Subject-Object Generation:

• This configuration allows you to input a reference subject image and a reference object image that will be combined in a generated scene according to your text prompt.

Bottom Group - Single-Subject Generation:

• This configuration uses a single reference image and generates new images of that subject in different contexts.

3.2 Generation Process with ByteDance's UNO

Basic Setup:

1. Load Reference Images:
   • For subject-object workflow (top group): Upload one subject reference image and one object reference image
   • For single-subject workflow (bottom group): Upload one reference image
2. Set Text Prompt for ByteDance's UNO:
   • Write a descriptive prompt for your desired scene, including how the reference subject/object should appear
3. Adjust ByteDance's UNO Parameters:
   • Width/Height: Determine output resolution (recommended 768×1024 or 1024×1024)
   • Guidance: Controls how closely ByteDance's UNO generation follows your text prompt (3.5-4.0 recommended)
   • Num_steps: Higher values (20+) produce more consistent results with ByteDance's UNO, especially for human faces
   • Seed: Set a specific seed for reproducibility or randomize for variety in ByteDance's UNO outputs
4. Generate Image with ByteDance's UNO:
   • Click "Queue Prompt" to run the ByteDance UNO workflow
   • Review the result in the "Save Image" node

3.3 Parameter Reference for ByteDance's UNO

UNO Loadmodel Node:

• model_type: Determines which ByteDance UNO model variant to use
  • flux-dev: Standard ByteDance UNO model with good balance of quality and performance
  • flux-schnell: Faster ByteDance UNO model variant with slightly lower quality

3.4 Tips for Best Results with ByteDance's UNO

Subject Selection for UNO:

• Choose clear, well-lit reference images where the subject is prominent
• For faces/people, front-facing images with neutral expressions work best with ByteDance's UNO
• For objects, clear images with minimal background distraction are ideal for ByteDance's UNO processing

Prompt Engineering for UNO:

• Be specific about subject descriptions that match your reference images
• Describe the desired scene, lighting, and composition for ByteDance's UNO to generate
• Include reference to the specific subjects (e.g., "a young man", "a vintage bicycle")
• Keep prompts concise but detailed for optimal ByteDance UNO results

4. Technical Information about ByteDance's UNO

The ByteDance UNO technology is built upon FLUX models and leverages the DiT (Diffusion Transformer) architecture with several key enhancements:

• Universal Rotary Position Embedding (UnoPE): A specialized mechanism in ByteDance's UNO that helps the model differentiate between reference subjects and objects, significantly reducing attribute confusion.
• Progressive Cross-Modal Alignment: Training methodology that enables ByteDance's UNO to handle increasingly complex multi-condition controls.
• Universal Customization: ByteDance's UNO is capable of handling diverse subject types without specialized training.

5. Acknowledgements

This workflow is powered by UNO, developed by the Intelligent Creation Team at ByteDance (Shaojin Wu, Mengqi Huang, Wenxu Wu, Yufeng Cheng, Fei Ding, and Qian He). The ComfyUI integration of ByteDance's UNO is provided by community contributors, adapting ByteDance's groundbreaking research into an accessible workflow.

For additional information about ByteDance's UNO, visit the .