Bernini: ByteDance's Semantic-Planning Video Framework
Bernini is ByteDance's open video generation and editing framework: a language-model semantic planner plus a Wan2.2-based video renderer, released under Apache-2.0.
Bernini by ByteDance
Video GenerationVideo EditingDiTByteDanceBernini is ByteDance's open video generation and editing framework. A multimodal language model acts as a semantic planner that writes a shot plan, and a diffusion transformer renderer turns that plan into video. The same pipeline covers text-to-video, reference-to-video, video editing, and aesthetic transfer, and it ships in diffusers format for ComfyUI.
| Developer | ByteDance |
| Architecture | Semantic planner (multimodal LLM) + DiT video renderer on a Wan2.1/2.2 backbone |
| License | Apache-2.0 |
| Generation Modes | Text-to-video, reference-to-video, video editing, aesthetic transfer |
| Precisions | fp16, fp8, int8, MXFP8, NVFP4 (per release and per ComfyUI pack) |
Variants
| Variant | Description | License | Released |
|---|---|---|---|
| Bernini-R | DiT video renderer component: MLLM semantic planner plus a Wan2.1/2.2 renderer for generation and editing | Apache-2.0 | 2026-06 |
| Bernini v2 | Full semantic-planning pipeline in diffusers format: Qwen2.5-VL planner, Wan2.2 dual-DiT renderer, stronger reference-guided editing | Apache-2.0 | 2026-08 |
Select a variant above for curated weight downloads, tutorials, and related information.
What Bernini does
- Semantic planning: a multimodal language model converts the prompt and any reference images into an explicit shot plan before rendering, which keeps multi-subject scenes and edits on target.
- One pipeline, many tasks: text-to-video, reference-to-video, video-to-video editing, reference-guided editing, and aesthetic transfer all run through the same planner and renderer.
- Diffusers-first releases: ByteDance publishes the pipeline in diffusers format, so it loads with standard tooling and is repackaged for ComfyUI by the community.
ComfyUI support
Bernini is available in ComfyUI through community repackaging rather than a first-party node pack. The Bernini-v2-ComfyUI pack provides FP8 and NVFP4 weights and a single workflow covering the six supported tasks.
Comments
Sign in with GitHub to join the discussion.