tencent-WeVisCut-GGUF

WeVisCut is a 9B tool-using video-editing agent from Tencent, built on Qwen3.5-9B, that converts knowledge articles and narration scripts into multi-minute edited videos. Rather than generating videos directly, it uses 12 tools for planning, narration, asset retrieval, composition, inspection, and repair, with a separate runtime executing editing workflows. Trained on executable editing trajectories, it also supports reinforcement learning. Its Studio UI provides an editable timeline, manual and chat-based revisions, per-shot rerendering, pause/resume, and version rollback. The model card provides no quantitative benchmark results; output quality depends on configured retrieval, TTS, rendering services, and templates. BF16 inference requires approximately 24 GB GPU memory and Python 3.11+. The runtime is Apache-2.0, while the model weights carry an other license and separate usage terms. Intended for research and production assistance, generated videos should be reviewed for factual accuracy and quality before publication.

[!NOTE] MTP — Speculative Decoding: No

Model Files

llama.cpp

LLM inference in C/C++ — https://github.com/ggml-org/llama.cpp