📌 Optimal Setup Guidelines

📝Overview

This ComfyUI workflow seamlessly combines WD14, JoyCaption, and Ollama to generate detailed and accurate captions from images. While it is not 100% accurate, it gets the job done well for most cases. Some images may require running the workflow 2-3 times to achieve the best caption results.

📦 Required Packages - To ensure seamless functionality, please install the following ComfyUI custom nodes:

1. ComfyUI-WD14-Tagger

2. ComfyUI-Ollama

3. ComfyUI-Prompt-Reader-Node

4. ComfyUI-JoyCaption

5. Python Interpreter Node

6. ComfyUI-Various

7. Dolphin3

8. Two more inside zip file

⚙️ How to Use: Key Nodes Explained

This workflow centers around three main nodes that control captioning and tag processing. Here's how to use each:

1. JoyCaption Instruction

• Purpose: Generates initial image captions using the JoyCaption node.

• Usage:

  • Feed the image input into this node.

  • Adjust style or detail level if available in the node parameters to customize caption output.

  • The generated captions serve as a base for further refinement.

2. Ollama Instruction

• Purpose: Refines and enhances captions by leveraging Ollama’s language model.

• Usage:

  • Take the output text from JoyCaption as input here.

  • Use Ollama Instruction to add context, clean up, or expand on the caption text.

  • You can tweak prompt templates or parameters to better match your desired caption style.

3. Remove Tags

• Purpose: Cleans up unwanted or irrelevant tags from generated captions to improve prompt quality.

• Usage:

  • You can configure the tag list to remove certain words or phrases.

  • Ensures the final caption is concise and useful for downstream tasks (e.g., text-to-image generation).

Note:

• Depending on the instructions provided to these nodes, the output may vary greatly. Experiment with different prompts and parameters to achieve the best results for your use case.

• The current instructions are optimized to work well for both SFW (safe-for-work) and NSFW (not-safe-for-work) content.