Sign in

Nodes Browser

ComfyDeploy: How z-a1111-sd-webui-DanTagGen works in ComfyUI?

What is z-a1111-sd-webui-DanTagGen?

Improved DanTagGen implementation that posesses all functionality of the A1111 webui extension.

How to install it in ComfyDeploy?

Head over to the machine page

  1. Click on the "Create a new machine" button
  2. Select the Edit build steps
  3. Add a new step -> Custom Node
  4. Search for z-a1111-sd-webui-DanTagGen and select it
  5. Close the build step dialig and then click on the "Save" button to rebuild the machine

ComfyUI node better aligned to KBlueLeaf's implementation of DanTagGen (https://github.com/KohakuBlueleaf/z-a1111-sd-webui-dtg)

Features:

  • Able to change DTG model, requires Tipo-KGen (https://github.com/KohakuBlueleaf/KGen)
  • Able to change seed, temperature, as well as top_p & top_k (top_p & top_k were not implemented in other custom nodes for DTG) image

A1111-SD-WebUI-DTG

A sd-webui extension for utilizing DanTagGen to "upsample prompts".

It can generate the detail tags/core tags about the character you put in the prompts. It can also add some extra elements into your prompt.

an extra z is added to repo name to ensure this extension will run process() after other extensions

What is DanTagGen

DanTagGen(Danbooru Tag Generator) is a LLM model designed for generating Danboou Tags with provided informations. It aims to provide user a more convinient way to make prompts for Text2Image model which is trained on Danbooru datasets.

More information about model arch and training data can be found in the HuggingFace Model card:

KBlueLeaf/DanTagGen-beta · Hugging Face

How to use it

After install it into the sd-webui or sd-webui-forge. Just enable it in the acordion. It will automatically read the content of your prompt and generate more tags based on your prompt.

Options

  • tag length:
    • very short: around 10 tags
    • short: around 20 tags
    • long: around 40 tags
    • very long: around 60 tags
    • short or long is recommended
  • Ban tags: The black list of tags you don't want to see in final prompt. Regex supported.
  • Prompt Format: The format of final prompt. Default value is the recommended format of Kohaku XL Delta
  • Seed: the seed of prompt generator. Since we use temperature/top k/top p sampling, so it is not deterministic unless you use same seed. -1 for random seed.
  • Upsampling timing:
    • After: after other prompt processings, for example: after dynamic prompts/wildcard.
    • Before: Before other prompt processings.
  • Temperature: Higher = more dynamic result, Lower = better coherence between tags.

Example

image

Faster inference

You can now select models in Generation config accordion. The script will automatically download the selected models if it is not existed. (Include gguf model.)

If you want to download the gguf model by yourself, you need to rename it to <Model Name>_<GGUF Name> for example: DanTagGen-gamma_ggml-model-Q6_K.gguf