1. Workflow Overview

• Purpose: Generates high-quality images with mechanical bionic aesthetics (e.g., robotic insects, futuristic armor), suitable for concept art or sci-fi visuals. Supports bilingual prompts.

• Core Models:

  • 高级质感人像-FLUX_v1.0: Base model for high-detail rendering.

  • 机械仿生科技_D1.05 LoRA: Adds gear/mechanical textures (weight=1.0).

  • BaiduTranslateNode: Auto-translates Chinese/English prompts.

2. Key Nodes & Installation

1. CheckpointLoaderSimple

   • Function: Loads the base model (Download Link).

2. LoraLoader

   • Function: Applies the 机械仿生科技_D1.05 LoRA (Download Link).

3. BaiduTranslateNode

   • Install: Requires pysssss custom node.

3. Workflow Structure

Group 1: Online Mode (Default)

• Flow:

  1. Text input → Translation → CLIP encoding + LoRA fusion.

  2. Image generation via Euler sampler (30 steps, CFG=7).

• Output: 768x1024 images with mechanical textures.

Group 2: Local Mode (Manual Enable)

• Optimizations:

  • VRAM control: EmptyLatentImage for batch size.

  • Quality: VAE-ae.sft for detail enhancement.

Group 3: Style Control

• Key Nodes:

  • BasicGuider: Adjusts style intensity.

  • LoRA: Forces gear/metal elements.

4. Inputs & Outputs

• Required Inputs:

  • Prompt (e.g., "Metallic exoskeleton with blue mechanical details").

  • Resolution: Default 768x1248 (adjustable in local mode).

• Outputs:

  • HD mechanical bionic image (PNG).

  • Text description (optional via ShowText).

5. Notes

• VRAM:

  • Online: ≥8GB GPU.

  • Local: ≥12GB GPU (use --medvram).

• Troubleshooting:

  • Over-stylization: Reduce LoRA weight (current 1.0 → 0.6-0.8).

  • Translation errors: Check BaiduTranslateNode API config.

• Tips:

  • Use Simple scheduler for faster runs.

  • Disable unused modules (e.g., translation) to save resources.