1. Workflow Overview

This workflow automatically enhances muscle definition in portraits using AI. Key features:

• Muscle Enhancement: LoRA models refine muscle details.

• Local Inpainting: Masks control muscle area generation.

• Preprocessing: Depth maps, resizing, etc.

Core Models:

• Base Model: 基础算法_F.1 (UNET for image generation).

• LoRAs:

  • 肌肉身材_v1.0 (muscle enhancement).

  • 黑森林_Depth&Canny Lora (depth/edge control).

  • XLabs F.1 Realism LoRA_V1 (realism boost).

• Auxiliary Models:

  • DepthAnythingV2 (depth map generation).

  • SegmentAnythingUltra V2 (auto-segmentation).

2. Key Nodes & Installation

Dependencies:

• Place LoRAs in models/loras.

• Download depth models from Hugging Face/Civitai.

3. Workflow Structure

Group 1: Input & Preprocessing

• Input: Portrait (e.g., 64da76e6817701...jfif).

• Steps:

  1. Load image via LoadImage.

  2. Resize with ImageScale (e.g., 512x512).

  3. Generate depth map via DepthAnythingV2.

Group 2: Muscle Generation

• Input: Preprocessed image, depth map, prompts (e.g., "Muscles, a strong body").

• Steps:

  1. Encode prompts with CLIPTextEncode.

  2. Apply muscle LoRA (strength 0.72).

  3. Generate enhanced image via KSampler.

Group 3: Inpainting & Blending

• Input: Generated image, original, masks.

• Steps:

  1. Segment body with SegmentAnythingUltra.

  2. Refine masks via GrowMaskWithBlur.

  3. Blend results with ImageComposite+.

Group 4: Output

• Result: Saved as PNG in ComfyUI/output.

4. Inputs & Outputs

Inputs:

• Image: Full-body/half-body portrait.

• Prompts: Include "muscle," "bodybuilding," etc.

• LoRA Strength: Adjustable (e.g., default 0.72 for 肌肉身材_v1.0).

Output:

• Enhanced PNG image.

5. Notes

1. VRAM: ≥12GB recommended (multi-model inference).

2. Mask Accuracy: Tweak SegmentAnythingUltra prompts (e.g., "arm, chest") if needed.

3. Compatibility:

   • Some LoRAs require SDXL.

   • DepthAnythingV2 needs CUDA.

4. Optimization:

   • Disable live preview (VHS_latentpreview=false).

   • Launch ComfyUI with --highvram.