Open-source Deepbooru TagWalker cleans LoRA dataset tags one keypress at a time
Waka-Neko · reddit · 2026-09-21
Developer Waka-Neko released Deepbooru TagWalker, a free open-source tool for cleaning LoRA training dataset tags. Instead of editing caption files one by one, it flips the problem: you judge one tag across every image that has it — Yes, No, next — with each decision written to disk instantly, plus custom filters for batch actions.
Key features:
- Tag knowledge: in-program Danbooru lookups across four eras (2017, Pony, Illustrious/NoobAI, current) show if your base model even knows a tag, with near-synonym compare mode; works offline.
- Real-time token counting against the tokenizer your trainer actually uses — CLIP for SDXL/Illustrious, T5 for Flux.1, Qwen3 for Flux.2 Klein, Mistral for Flux.2 Dev — avoiding silent truncation that degrades training.
- Dataset health checker flags missing/empty captions, odd resolutions, stray files; stats can be exported to an LLM for pruning advice.
- Safe bulk edits: undoable rename/split/delete/dedup across the whole set, underscore/space conversion, conflict rules catching contradictory captions.
- Built-in image editor for cropping to Kohya-style training buckets.
GitHub: Elliezrah/deepbooru-tagwalker
More from Multimodal
- Cinematic Seedance 2.5 AI video clip goes viral on X without any prompt context — CurieuxExplorer · 2026-09-21
- Runway Big Pitch Contest entry 'ALL THE WATER' imagines oceans vanishing underground — bennash · 2026-09-21
- MiniMax H3 Video: How to Use a Second Image to Replace a Body Part — Next-Place0 · 2026-09-21
- Paper: Physically Based Rendering in the Latent Space — ssh4net · 2026-09-21
- Adaptive Color Grading paper: KNN beats end-to-end models at tonescale prediction — ssh4net · 2026-09-21
- Blender as Director, Seedance as Renderer: A Cinematic AI Video Workflow — CurieuxExplorer · 2026-09-21