ComfyUI.org

Unlock Precise AI Image Editing with Flux Diffusion and Multi-Condition Guidance

CN
ComfyUI.org
2025-04-09 07:18:02

1. Workflow Overview

m99lke8iqq2ooa5ny63f5d722295a7ed0fd1a1b566ba89fa746fec28807d062fa99aaa8972a66fde6e.png

This workflow enables precise AI image editing via text instructions, leveraging Flux diffusion and multi-condition guidance to:

  • Modify details: Change hair color, adjust makeup, add accessories.

  • Transfer styles: Convert photos to illustrations or vintage tones.

  • Upscale resolution: Auto-adjust to SDXL standards (e.g., 1024x1024).

2. Core Models

Model Name

Description

F.1-BaseModel_fp16

Flux-optimized UNet for high-precision edits.

ae.sft

Custom VAE for enhanced color accuracy.

t5xxl_fp16+clip_l

Dual CLIP encoders for bilingual prompt understanding.

3. Key Nodes & Installation

  • FlowEditGuider

    • Function: Balances original (source_cond) and edit prompts (target_cond) dynamically.

    • Tuning: Adjust weights via FluxGuidance (default: 2.5/3.0).

  • BaiduTranslateNode

    • Function: Translates Chinese prompts to English (requires Baidu API key).

    • Install: Get ComfyUI-BaiduTranslate via ComfyUI Manager.

  • ImageGenResolutionFromImage

    • Function: Auto-calculates SDXL-optimal resolution (e.g., 816x1088 β†’ 1024x1360).

  • SamplerCustomAdvanced

    • Function: Skips initial steps (skip_steps) to preserve original features.

4. Workflow Structure

Group Name

Description

Input Image

Load source image, resized to 1024px on the longer edge.

Source Prompt

Auto-generated (WD14) or manual input (e.g., "woman with blonde hair").

Edit Prompt

Specify changes (e.g., "change hair to pink, add gold glasses").

Flux Control

Tune guidance strength and skip steps for edit fidelity.

Output Compare

Side-by-side original/edited results (slider supported).

5. Inputs & Outputs

  • Inputs:

    • Source image (PNG/JPG, β‰₯800px recommended).

    • Source prompt (auto or manual).

    • Edit prompt (must partially align with source).

  • Output:

    • Edited HD image (PNG).

    • Comparison view (via Image Comparer).

6. Notes

  • VRAM: β‰₯12GB GPU recommended; reduce resolution for complex edits.

  • Prompt Tips:

    • Preserve key features (e.g., pose/background) in edit prompts.

    • Use "keep XX" syntax (e.g., "keep red dress, only change background").

  • Troubleshooting:

    • Increase skip_steps (e.g., 10β†’20) if artifacts occur.

    • Ensure VAE is ae.sft for color accuracy.

previous
Unveiling the Past: Transforming Ancient Paintings into Hyper-Realistic Photos
next
Unleash the Power of WanVideo: Create Stunning Sticker Peeling Effect Videos

Recommend