The checkpoint is trained for instruction-guided video editing with complex prompts. It combines a Wan-based video editing backbone with the corresponding connector weights for Qwen3-VL features.
💻 How to Use
Please refer to the code repository for environment setup, weight download, training, inference, and evaluation instructions.
By default, the released code expects the checkpoint at:
weights/ThinkV2V/thinkv2v_5b.safetensors
You can also provide an explicit checkpoint path through THINKV2V_CKPT when running inference or continued training.
🎯 Intended Uses
This checkpoint is intended for research on instruction-guided video editing. Users should evaluate generated outputs carefully before applying the model in downstream workflows.
⚖️ License and Ethics
The checkpoint is released under the Apache License 2.0.
Users should apply the model responsibly and avoid harmful, deceptive, privacy-violating, or otherwise inappropriate use. Generated videos are determined by the input video, text instruction, model weights, and runtime settings.
