ComfyUI.org

Unlock Dark-Style Image Generation with AI-Powered Hybrid Workflow

CN
ComfyUI.org
2025-06-24 09:08:26

1. Workflow Overview

mcaaz0tqi7zxd2t7upkyasuoban.png

This is a dark-style image generation + auto-captioning hybrid workflow, featuring:

  • Automatic image captioning via Florence-2-large model

  • Dark-style image generation using custom @迁士-F.1 ancient-style model

  • Supports both Text-to-Image and Image-to-Image modes

2. Core Models

Model Name

Function

Source/Installation

@迁士-F.1 Human Model

Ancient dark-style generation

Manual .safetensors download

Florence-2-large

Image-to-caption generation

Auto-loaded via Florence2ModelLoader

LoRA: undefined_0003

Style fine-tuning (weight=0.8)

Bundled in workflow

3. Key Nodes

Node Name

Function

Installation

Florence2Run

Generates captions from images

Install ComfyUI-Florence2 plugin

SamplerCustomAdvanced

Custom sampler (Euler, etc.)

Built-in

DualCLIPLoader

Loads dual CLIP models

Built-in

4. Workflow Groups

  • Group 1: Captioning

    • Input: Uploaded image (via LoadImage)

    • Output: Text caption (e.g., "A black-and-white photograph of...")

  • Group 2: Text-to-Image

    • Input: Trigger word (e.g., "dark wind") + generated caption

    • Output: 1024x1536 dark-style image

5. Inputs & Outputs

  • Input Parameters:

    • Required: Image (Caption mode) / Trigger word (Text mode)

    • Optional: Seed (random by default), Resolution (1024x1536)

  • Output:

    • Generated image (PNG with metadata)

    • Caption text (JSON format)

6. Notes

  • Must download microsoft/Florence-2-large (~5GB)

  • VRAM: ≥12GB (FP8 precision required)

  • Troubleshooting: If LoRA load failed, check undefined_0003.safetensors path

previous
Manga/Novel Video Auto-Repaint Tool
next
Edit Model Poses Like a Pro: A Step-by-Step Flux Kontext Workflow

Recommend