ByteDance open-sources Bernini video framework

By decoupling the structural planning of edits from the actual pixel generation, Bernini addresses the fundamental flaw of temporal inconsistency that plagues modern video diffusion models, proving that semantic intelligence is the prerequisite to reliable video manipulation.

• –**Architectural Shift:** Moving away from direct pixel manipulation to feature-space semantic planning allows for much finer control over editing decisions.
• –**Practical Consistency:** The ability to lock unaffected regions makes it viable for professional production lines where consistency is non-negotiable.
• –**Accessible Tooling:** Releasing the framework and model under Apache-2.0 fosters rapid developer adoption and integrations, particularly within ComfyUI pipelines.