LoRA caption helper
Inconsistent captions confuse training. Paste lines or comma blobs, strip extra spaces, remove duplicates, and enforce underscore style for Danbooru-like datasets. Export a single line you can drop into a caption file or metadata slot.
Try it now
Runs in your browser. AI-assisted steps need a free account, credits where applicable, and daily limits to keep latency fair for everyone.
Why duplicates hurt
Repeated tags shift weighting. The helper collapses duplicates while preserving order.
Pairs with Anifusion model training
When you upload packs, clean captions here first for faster review.
Batch-friendly workflow
Process one file at a time in long sessions, or split huge dumps so the tab stays responsive. The goal is predictable tokens per image so your training job does not overweight repeated typos or stray double spaces.
Related links
FAQ
Does this call an API?
No. All normalization is local.
Can I keep natural language captions?
Use a lighter mode: only trim whitespace without forcing underscores.
Is there a tag limit?
Very large pastes may slow your browser. Split files if needed.
Does it translate tags?
No translation. It formats text you provide.