Vchitect/ShotVL-7B
Vchitect/ShotVL-7B is a 7 billion parameter vision-language model fine-tuned from Qwen2.5-VL-7B-Instruct, developed by Vchitect. It specializes in expert-level cinematic understanding, trained on a large, high-quality dataset for this domain. The model achieves state-of-the-art performance on the ShotBench benchmark, excelling at tasks like identifying shot size, framing, camera angle, and movement in both images and videos, with a context length of 32768 tokens.
Loading preview...
ShotVL-7B: Expert-Level Cinematic Understanding
ShotVL-7B is a 7 billion parameter vision-language model developed by Vchitect, fine-tuned from Qwen2.5-VL-7B-Instruct. This model is specifically designed for expert-level cinematic understanding, leveraging a large and high-quality dataset focused on cinematic language. It processes both image and video inputs, making it highly versatile for film analysis tasks.
Key Capabilities
- Advanced Cinematic Analysis: Excels at identifying intricate cinematic elements such as shot size (SS), shot framing (SF), camera angle (CA), lens size (LS), lighting type (LT), lighting conditions (LC), shot composition (SC), and camera movement (CM).
- State-of-the-Art Performance: Achieves new state-of-the-art results on the comprehensive ShotBench benchmark, demonstrating superior accuracy compared to other open-sourced and proprietary VLMs.
- Multimodal Input: Capable of analyzing both still images and video sequences to extract cinematic information.
Good For
- Film and Media Analysis: Ideal for researchers, critics, and students studying cinematography and film language.
- Automated Content Tagging: Can be used to automatically tag and categorize video content based on cinematic properties.
- Creative Production Tools: Potentially useful in film production workflows for pre-visualization or post-production analysis.
- Benchmarking Cinematic AI: Serves as a strong baseline and SOTA model for evaluating new vision-language models in cinematic understanding.