Tasks and inputs
WeVisCut is listed for Multimodal, Video-to-video. It accepts text and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
WeVisCut is Tencent's 9B tool-calling agent that edits narration scripts into multi-minute videos through 12 editing tools and an editable Studio timeline.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Open weights; no hosted price. Your TTS, retrieval and rendering services carry their own costs.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
WeVisCut does not generate pixels. Tencent's 9B model is an editor: given a knowledge article or narration script, it decides, one tool call at a time, how to turn it into a multi-minute cut. Segment the article, plan shots, synthesise narration, fetch footage, fill animated templates, choose music, judge the render, repair what failed, stitch. About twelve calls per rollout, chosen by the policy itself rather than a hard-coded stage machine.
| Repository | tencent/WeVisCut |
| Base | Qwen3.5-9B |
| Uploaded | 9 September 2026 (project announcement and report in October) |
| Action space | 12 editing tools |
| Runtime | WeVisCut Studio |
| Requirements | Python 3.11+, ~24 GB NVIDIA GPU for bf16 |
Licensing needs a careful read. Tencent's code is Apache 2.0, but the weight download is marked with a separate licence and says the weights and third-party components follow their own terms. Check the weight terms specifically before shipping videos commercially.
Each step, the policy reads a structured editing state (chapters, shots, narration, footage, templates, music, render) and emits exactly one call. Guardrails can reject an illegal call but never substitute their own choice. Quality gates are tools too: a text judge, a vision-language judge on rendered video, and a self-reflection step that decides what to fix next. Narration synthesis locks each shot's duration, so timing follows the voice track.
The Studio keeps the result editable. Eight synchronised tracks, pause and resume, manual edits, chat-based revisions, per-shot re-rendering and version rollback all come with it, and there is a UI-only mode that runs without loading weights.
Output quality depends heavily on what you plug in: retrieval sources, text-to-speech, renderer and templates. The policy coordinates; it cannot rescue poor stock footage. Tencent trained it on more than 130,000 examples reconstructed from 3,750 finished videos by "editing inversion" and verified replay, and reports that it matches Claude Opus 5 under the same tools while generating nearly 33 times fewer tokens per successful output minute. That is a publisher claim, measured on Tencent's own 100-task benchmark.
Take five articles you have already turned into videos by hand, run them through the Studio with your own TTS and footage sources, and count how many revisions each draft needs before it is publishable. Also review narration for factual drift; Tencent explicitly does not guarantee accuracy. If you need generated footage rather than edited stock, look at video generators such as Kling VIDEO 4.0 or Seedance 2.5 as upstream sources. Other Tencent releases are on the Tencent provider page.
Compare the model on the work you need to complete.
WeVisCut is listed for Multimodal, Video-to-video. It accepts text and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.
The weights are downloadable under Apache-2.0 code; weights under their own terms. Check runtime support and size the complete model and working context for your expected concurrency.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
NVIDIA
Cosmos3-Nano-Sim-Auto is NVIDIA's open causal driving simulator: real-time multi-camera video and LiDAR that react to an AV policy's planned path.
NVIDIA
Cosmos3-Nano-Sim-Transfer is NVIDIA's open 16B model that renders simulator depth, segmentation or edge maps into photorealistic robot video in real time.
NVIDIA
Cosmos3-Nano-Transfer-Auto is NVIDIA's open 16B model that generates synchronised multi-camera driving video and LiDAR from map controls and a prompt.
skywork
Matrix-Game 4.0 is Skywork's open set of 5B real-time video renderers that turn a running game's frames into clean styled video, per scene.
skywork
SkyReels-V3-V2V-14B is Skywork's 14B video-extension model that continues a clip for 5-30 seconds or cuts to a new shot, at 720p from a text instruction.
AInVFX
AInVFX/SeedVR2_comfyUI is a source-linked Hugging Face repository assigned to Video-to-video. Its recorded task interface accepts video and produces video.