Unlock MidJourney-Style Image Generation with Multi-LoRA Fusion Workflow
1. Workflow Overview
This workflow replicates MidJourney-style image generation by combining multiple LoRAs and precise prompt control. Key features:
Multi-LoRA fusion (e.g., MJv6, MJ52 for art styles).
Auto captioning (generates prompts from input images).
Negative prompt filtering (excludes low-quality/deformed content).
2. Core Models
Model Name | Function |
|---|---|
| Base model for 2D illustration style. |
| MidJourney V6 style LoRA (modern art enhancement). |
| Detail-enhancing LoRA. |
| Adds textures/fine details. |
3. Key Nodes
3.1 Required Custom Nodes
ComfyUI-Custom-Scripts: Provides
ShowTextandStringFunctionnodes (install via Manager).ComfyUI-LoRA-Stack: Supports multi-LoRA stacking (manual GitHub install).
Miaoshouai_Tagger: Auto captioning (download
promptgen_large_v2.0tomodels/taggers).
3.2 Dependencies
LoRA Files: Place
mjv6.safetensorsetc. inmodels/loras.Negative Prompts: Built-in filters (e.g.,
"low quality, deformed, watermark").
4. Workflow Structure
Group Name | Inputs | Outputs | Logic |
|---|---|---|---|
Image Input | Uploaded image | Caption text | Generates prompts via Tagger. |
LoRA Stack | Base model + 3 LoRAs | Fused model | Blends MJv6/MJ52 styles. |
Text Encoding | Positive/negative prompts | Conditioning vectors | Controls content and style. |
Generation | Resolution (512x512), seed, sampler | Latent image | Uses DPM++ 2M sampler. |
Post-process | Latent + VAE | Final PNG | Decodes and previews results. |
5. Inputs & Outputs
Inputs:
Required: Reference image (e.g.,
71Fd7Zowg3L._AC_SL1500_.jpg), base model.Optional: Seed (random by default), LoRA weights (e.g., MJv6 weight 0.8).
Output: Generated image (MidJourney-style artwork).
6. Notes
LoRA Conflicts: Enable LoRAs step-by-step to avoid style clashes.
VRAM: ≥8GB recommended when loading multiple LoRAs.
Prompt Tuning: Manually refine auto-generated captions from
Miaoshouai_Tagger.