TextVerifier catches flawed Chinese glyph rendering with F1 0.844, far above OCR and VLMs
ATH-MaaS · hf · 2026-10-15
Researchers from PKU, Alibaba, and Zhejiang University propose TextVerifier, which verifies Chinese text rendering in generated images — catching local defects like extra/missing strokes that OCR and vision-language models are blind to due to their recognition-only training.
The verifier feeds the text tower a character's Ideographic Description Sequence (IDS) instead of a single glyph, trained on 12.7M crop-IDS pairs plus label-free "same-character, different-quality" hard pairs. They also release TEXTRENDERBENCH-ZH, the first per-character benchmark for generated Chinese text (7,358 annotated characters). TextVerifier reaches F1 0.844 vs 0.526 for OCR confidence, 0.329 for a general VLM, and 0.590 for TextPecker; used as a per-character reward for Flow-GRPO post-training it further reduces rendering errors.
More from Multimodal
- Fish Audio's Drama-3 voice-cloning Space trends on Hugging Face — fishaudio · 2026-10-15
- Midjourney v8.2 style test: bold flat-ink illustration looks, sref code included — azed_ai · 2026-10-15
- Sarvam AI launches Bulbul V4 Flash, a faster compact TTS model for Indian languages — CurieuxExplorer · 2026-10-15
- Maya-2 debuts as open-weight Indic voice model, ranks #2 on Hindi TTS leaderboard — CurieuxExplorer · 2026-10-15
- PixText fixes text in generated images: 83% glyph pass rate vs 37% for baselines — ATH-MaaS · 2026-10-15
- AI Video Creators: 'Amazing Visuals Are Solved — Unique IP and Characters Are What's Left' — AIandDesign · 2026-10-15