SOURCE-REVIEWED MODEL PROFILE
WeVisForensics-4B adapters
Five LoRA adapters for an agentic video-forensics workflow, dependent on Qwen3.5-4B.

Metadata reviewed 2026-11-20 · Original creator: Tencent
What it does
- Main investigator and four visual inspector adapters
- Rank 64 / alpha 128 LoRA configuration
- Separate speech, speaker and OCR tool dependencies
Local fit
Publisher Transformers demo guidance:GPU at least24GB. Adapter repository is not a full standalone checkpoint. Third-party tools retain their own licenses; detector output is not an authenticity verdict.
Base model context is not an independently verified agent-workflow limit.
Evidence and limits
Source reviewed; no local inference, forensic evaluation or hardware benchmark.
Publisher workflow documented; not independently tested
License and source
Published artifact LICENSE is Apache-2.0; separate dependencies retain their own terms.
Read the original model card ↗
Reviewed source revision: 4281c2167705a48f84334026f40bba394d461067