1. Workflow Overview

This is an advanced holographic tech-style image translation workflow featuring:

• Depth-aware image repainting (DepthAnything V2)

• FLUX-based futuristic holographic effects

• UltimateSD upscaling integration

• Multi-stage conditioning control

Core Models:

• DepthAnything V2: Depth map generation

• F.1_Depth-fp16: FLUX base model with depth control

• Holographic LoRA: Style fine-tuning (weight 0.8)

• R-ESRGAN_4x+ Anime6B: Upscaling model

2. Critical Nodes

Dependencies:

• Model Files:

  • Place ae.sft in models/vae

  • Download 全息科技素材_V1.0.safetensors to models/loras

• Plugins: Essential to use ComfyUI-Manager

3. Workflow Structure

Key Groups:

1. Input (Left):

   • Load image → Resize to 1024x1024 → Generate depth map

   • Prompt: "quanxi,3D rendering,holographic"

2. Main Process (Center):

   • Depth map + FLUX generation → 30-step Euler sampling

3. Upscale (Right):

   • 2x UltimateSD upscale → 512x512 tile processing

4. Inputs & Outputs

Required Inputs:

• Source image: PNG/JPG (e.g. 未标题-2.png)

• Resolution: Auto-adjusted to 1024x1024

• Seed: Randomizable (default)

Final Output:

• Format: PNG with metadata

• Pipeline: Original → Depth conversion → Holographic render → Upscale

5. Notes

• VRAM: ≥12GB required (more for upscaling)

• Common Errors:

  • Depth model failed: Verify depth_anything_v2_vitl_fp32.safetensors

  • LoRA conflict: Avoid multiple style LoRAs

• Optimization:

  • Reduce upscale tile size to 256x256

  • Use fp16 models for faster generation