Using closed-decision Jev-like models to kill tag hallucination in dataset captioning
Iory1998 · reddit · 2026-09-21
A Reddit user proposes replacing free-form LLM captioning with closed-decision 'Jev-like' models for training dataset tagging.
Problem: LLMs hallucinate out-of-vocabulary Danbooru tags and drift across batches, which breaks models like Illustrious and Anima that only respond to tags in their learned vocabulary.
Core idea: Invert open questions into closed ones (photo or illustration? hair color from a list? character count 0/1/2/3+?). Each model pass picks from a fixed state list with probabilities; only above-threshold tags enter the caption — every dimension gets answered and zero invented tags. To save compute, a cheap first pass shortlists candidates, then a Jev-like pass confirms each against the image. Finally, the frozen tag list is handed to an LLM purely as a translator to produce a natural-language description, so one captioning run yields both Booru tag strings and prose for either training architecture.
Caveat: it can still affirm a real but absent tag — bounded failure, with the threshold as your recall/precision dial. Runs in ComfyUI, no fine-tuned captioner or API. The author hasn't built it yet and invites feedback.
More from coding & agent
- If you ship AI code without reading it, your project doesn't matter — yacineMTB · 2026-09-21
- Teknium floats dedicated plugin pages in Hermes Agent plugins explorer — Teknium · 2026-09-21
- Yacine: not reading every line of core code makes you unserious — yacineMTB · 2026-09-21
- Agent builders say provider KV caching black boxes block swarm and long-run agents — Small_Luck8177 · 2026-09-21
- Jev Explained: Open-Source Repo With 4 Examples and an AI-Assisted Way to Learn Jev Primitives — leslysandra · 2026-09-21
- Open-Jev Released: Open-Source 2B/9B Decision Models with LoRA Adapters — BLUECOW009 · 2026-09-21