berniniai.org

Bernini AI is ByteDance’s unified framework for AI video generation and editing. It lets you generate video from text, restyle or modify existing footage using a prompt, and drive new clips from reference images.

Powered by an MLLM-based semantic planner and a DiT-based renderer, Bernini AI focuses on both reasoning about what should change and producing frames that preserve fine detail for editing tasks. It supports multimodal workflows so you can combine instructions with source video and up to five reference images.

Use it online in your browser: choose your task (text-to-video, reference-to-video, or prompt-based editing), upload JPG/PNG/WebP images as references, and generate or edit directly. When your clip is ready, download the result and iterate by adjusting your prompt and references.

What types of video tasks can Bernini AI do?

Bernini AI supports text-to-video, reference-to-video (driving a clip from images), and prompt-based editing of existing footage. It also includes image generation and image editing alongside video workflows.

How does reference-to-video work?

Reference-to-video lets you generate a new clip using up to five reference images to anchor a subject, object, or style. You can also combine references with text for more specific direction.

Can I edit existing video with Bernini AI?

Yes. Provide a source clip and describe the change in a prompt—Bernini AI is designed for targeted edits while aiming to keep the rest of the footage consistent.

Do I need H100 or multi-GPU hardware to use Bernini AI?

No for the online workflow. The platform runs the generation and editing in the cloud so you don’t need to set up a multi-GPU environment yourself.

What image formats and limits are supported for references?

You can upload JPG, PNG, or WebP files. Each image can be up to 10 MB.

Is Bernini AI open source?

Yes. Bernini is released under the Apache 2.0 license, with the inference code and renderer weights published on GitHub and Hugging Face for self-hosted use.