ByteDance unified video generation and editing framework pairing an MLLM semantic planner with a DiT renderer.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
Bernini is a research framework from ByteDance for video generation and editing, described in the paper "Bernini: Latent Semantic Planning for Video Diffusion" (arXiv 2605.22344). It combines an MLLM-based semantic planner, which reasons over text, source images, source videos and target placeholders, with a DiT-based renderer that performs flow-matching denoising over VAE latent tokens. A Segment-Aware 3D RoPE distinguishes tokens from different visual segments.
The project page demonstrates several tasks: prompt-driven video-to-video editing (V2V), reference-guided editing (RV2V) using images for object, material, weather or style guidance, content insertion of an image or video into a source video, and reference-to-video generation (R2V) with up to five reference images. According to a YouTube creator summary, it is built on Wan 2.2 with high- and low-noise branches of roughly 84 GB each, and a ComfyUI integration was reported to be in progress; these details are not confirmed on the project page.
The provided material does not include a code repository, weights, license or hosted demo link, so availability and terms are unverified. Hardware requirements appear large given the reported checkpoint sizes.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Open-source unified video editing model that edits videos via text prompts or image/video references (object removal, background swap, style change, reference insertion). Based on Wan 2.2 with high/low branches, ~84GB per branch; ComfyUI integration underway.
Closest neighbours by meaning, not just by category.
Model · ByteDance
ByteDance's open-weights video generation and editing model pairing a Qwen2.5-VL planner with a Wan2.2 renderer.
Model · ByteDance
ByteDance Seed's multimodal video generation model, producing video from text and reference inputs.
Model · ByteDance
ByteDance Seed's video generation model supporting multi-reference inputs and one-take creation.
Research · ByteDance
ByteDance research framework that distills multiple specialist image models into one via on-policy velocity matching.
Model · ByteDance
ByteDance research model unifying image and video understanding, generation and editing in a 3B active-parameter network.
Research · ByteDance
ByteDance and Texas A&M research method distilling image and video diffusion models into few-step generators adversarially.
