Youtu-Parsing-Omni
Youtu-Parsing-Omni is a 5B model that returns structured perception and description data from visual or audio input. Task prompts select the required output form.
Model design and access
The unified encoder handles images, audio and audiovisual video. Outputs can contain text, bounding boxes, reading order, timestamps, speakers or scene descriptions. The release supplies a vLLM plugin and fixed task prompts. Its custom license is not an unrestricted Apache grant.Youtu-Parsing-Omni official model documentation
Practical use
Practical guidance. Validate the schema and source alignment separately. A valid JSON result can still place text in the wrong region or assign speech to the wrong speaker. Preserve source coordinates and time ranges for review.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| License | Youtu-Parsing license; includes an EU territorial limitationYoutu-Parsing-Omni custom license |
| Output | Structured JSON with task-specific fieldsYoutu-Parsing-Omni official model documentation |
Related models
- HunyuanOCR-1.5
Tencent ยท HunyuanOCR
A lightweight document-parsing model with updated OCR training and optional DFlash decoding.
Launch date not confirmedDownloadable