1. Workflow Overview

This is an AI image generation and enhancement workflow based on the FLUX.1-dev model, optimized for architecture and interior design. Key features:

• Image Generation: High-quality design outputs from text prompts and reference images.

• ControlNet Integration: Uses depth maps, line art, and edge detection (Hed) for structural control.

• Face/Detail Enhancement: Automatic refinement via FaceDetailer.

• Upscaling: Tile-based processing (Tiled Diffusion) for high-resolution outputs.

• Multilingual Support: Vietnamese prompts auto-translated to English.

2. Core Models

3. Key Components

4. Workflow Structure

Group 1: Input & Preprocessing

• Inputs: Reference image, Vietnamese prompts.

• Outputs: Preprocessed images/text conditions.

• Key Nodes: LoadImage, JanusImageUnderstanding, BaiduTranslateNode.

Group 2: ControlNet Processing

• Inputs: Preprocessed images (e.g., depth maps).

• Outputs: Controlled generation parameters.

• Key Nodes: AIO_Preprocessor, ControlNetApplyAdvanced.

Group 3: Generation & Enhancement

• Inputs: Text conditions, ControlNet params, seed.

• Outputs: Initial generated image.

• Key Nodes: KSampler, FaceDetailer.

Group 4: Upscaling & Output

• Inputs: Low-res generated image.

• Outputs: HD image (up to 4K).

• Key Nodes: TTP_Image_Tile_Batch, ImageUpscaleWithModel.

5. Inputs & Outputs

• Inputs:

  • Reference image (drag-and-drop).

  • Vietnamese prompts (e.g., Hình ảnh chân thực).

  • Seed value (fixed/random).

• Outputs:

  • HD design image (PNG).

  • Intermediate results (e.g., ControlNet maps).

6. Notes

1. VRAM: Recommended 12GB+ (tile processing reduces usage).

2. Dependencies:

   • Manually download FLUX.1-dev and CTAI LoRA.

   • Install plugins: comfyui-impact-pack, janus-pro.

3. Compatibility: Requires ComfyUI v0.3.26+.

4. Optimization:

   • Enable TTP tiling for lower VRAM.

   • Adjust denoise (0.15~0.35) for speed/quality balance.