FilmGPT trains on feature films to edit raw footage into cinematic sequences
dyamins · x · 2026-07-23
FilmGPT presents a way to learn the grammar of film directly from feature movies. The system trains an autoregressive transformer on raw footage and then edits clips into coherent cinematic sequences, rather than generating new pixels.
- Venue: SIGGRAPH 2026.
- Core idea: learn film structure from movies themselves.
- Output: edited, coherent sequences from raw footage.
- This is a video-editing / cinematic understanding research project, not a generic text model.
More from Multimodal
- HeyGen’s HyperFrames adds a storyboard-first workflow for AI video generation — HeyGen · 2026-07-23
- Seedance 2.0 prompt share stitches GPT Image 2, Midjourney and Suno together — azed_ai · 2026-07-23
- Qwen3.6 35B multimodal GGUF build starts trending on Hugging Face — LuffyTheFox · 2026-07-23
- Nano Banana 2 Lite prompt turns a mecha collar into a cinematic macro portrait — fofrAI · 2026-07-23
- TwelveLabs previews Jockey, a video agent that works across an entire library in Claude — KadriJibraan · 2026-07-23
- MireloAI says its video-to-sound model generates synced SFX in seconds — aziz4ai · 2026-07-23