Tagger Prompt

Tagger Prompt is a small Forge extension that turns an input image into prompt tags inside the Stable Diffusion UI.

What It Does

• Adds a Tagger Prompt panel to the Forge interface.
• Lets the user upload an image or paste one from the clipboard.
• Runs the selected tagger model and generates tags.
• Supports filtering unwanted tags with Negative words.
• Supports optional Fooocus V2 Prompt Enhancement with adjustable Strength.
• Inserts the generated tags into the active prompt field.

Supported Taggers

• WD14
• WD3 (WD SwinV2 v3)
• WD ViT v3
• WD EVA v3
• WD Conv v3
• DeepDanbooru
• E621

Project Structure

• [scripts/tagger_prompt.py] contains the Forge/Gradio UI, settings, model auto-download logic, and the main user flow.
• [scripts/taggers_core.py] contains the ONNX-based tagger implementations.

Model Storage

• If tagger_prompt_models_dir is set in Forge settings, the extension uses that directory and expects the required model files to already be present there.
• If it is empty, the extension uses models/taggers_prompt_models.
• When the default models directory is used, the extension automatically downloads only the files required for the selected model on first use.
• If Fooocus V2 Prompt Enhancement is enabled, the extension also auto-downloads prompt-expansion assets on first use.

Prompt Controls

• Negative words: comma-separated words/phrases removed from generated tags.
• Fooocus V2 Prompt Enhancement: optional GPT-2 based prompt expansion.
• Strength (0.0 to 1.0): controls how much enhancement is applied.
  • 0.0 keeps original tagger output.
  • 1.0 applies full enhanced output.
  • Intermediate values blend original tags with added enhancement terms.

Typical Flow

1. Open the Tagger Prompt accordion in Forge.
2. Upload an image or paste one from the clipboard.
3. Select a tagger.
4. Wait for the tags to be generated.
5. Click Insert into Prompt.

Notes

• Models are loaded through onnxruntime.
• WD-style taggers support both general and character thresholds.
• DeepDanbooru uses only the general threshold.
• Fooocus V2 Prompt Enhancement requires transformers/torch at runtime.
• During first enhancement run, download progress is shown in console with [Fooocus V2] logs, and UI status reports whether files were downloaded or loaded from cache.

License

See LICENSE.md.