bytedance/Bernini

Health v2 · maintenance only

Flags: no_releases young

Adoption not part of the score

1287 stars · 100 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Bernini is a unified framework for video generation and editing that combines an MLLM-based semantic planner with a DiT-based renderer performing flow-matching denoising in VAE latent space. It includes open-sourced inference and training code plus model weights (1.3B and 14B renderer variants) released on HuggingFace.

Use cases

• generate videos from text prompts
• edit videos with natural language instructions
• remove watermarks or subtitles from videos
• apply style transfer to videos
• generate videos guided by reference images
• insert an image or video into an existing video
• train a custom video diffusion renderer

When to choose

• you need a unified open-source model for both video generation and instruction-based video editing
• you want reference-guided video generation with multiple input images
• you need DiT-based video diffusion weights compatible with the Diffusers ecosystem
• you want to reproduce or extend research on latent semantic planning for video diffusion

When to avoid

• you need lightweight CPU-only video processing without GPU hardware
• you only need simple video cutting or transcoding rather than generative editing
• you require a production-ready commercial-grade API with support guarantees
• your tasks are complex human generation where the smaller 1.3B model lags behind

Facets

library · maturity active

video-processing image-processing machine-learning deep-learning llm-inference image-processing artificial-intelligence deep-learning large-language-models python video-generation video-editing diffusion-transformer mllm text-to-video flow-matching diffusers bytedance video gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="bytedance/Bernini")