ByteDance / seedance 2.0 mini v2v

## Overview Seedance 2.0 Mini V2V is a video-to-video model from ByteDance. It uses a unified audio-video generation design. You give it reference footage plus a written edit brief. It re-renders a new clip that follows the source motion while applying your changes. The output can include synchronized sound, so you can ship a clip with audio in one step. ## What you can build - Product video edits that keep the same camera move but swap the product - Background and material changes, like water to lava, while motion stays consistent - Style transfer on existing footage, like “anime” or “film noir,” without re-cutting - Social variants in vertical, square, or widescreen from the same reference clip - Music- and rhythm-led cuts by guiding generation with an audio reference ## Inputs - One to three reference videos in MP4 format. Some providers also accept MOV. The combined reference duration must stay within a short-clip range, typically 2–15 seconds. Keep the total upload size within the provider’s cap. - An optional written prompt that describes what to change and what to keep. You can point to specific references inside the prompt using tokens like “Video 1” or “Image 1,” depending on the host. - Optional reference images in JPG, PNG, or WebP. You can usually include up to 9 images. Each image must stay under the host’s file size cap. - Optional reference audio in MP3 or WAV. You can usually include up to 3 clips, with total duration capped around 15 seconds. If you add audio guidance, at least one image or video reference is typically required. - A required output resolution choice. Mini tiers commonly support 480p and 720p. - A required aspect ratio choice. Common options include 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16. Some hosts also offer an auto mode that picks based on your inputs. - A required output duration in seconds within a short-clip window. Many hosts allow an auto duration mode. - An audio toggle that controls whether the output includes generated synchronized audio. ## Outputs The model returns a generated MP4 video. The video contains the edited scene rendered from your reference clip and prompt. When audio is enabled, the MP4 includes synchronized sound, which may include ambience, effects, and voiced dialogue from quoted lines. Some hosts also return a numeric seed value with the result for traceability. ## Recommended settings - Use 480p for quick drafts and iteration. Switch to 720p for review-ready outputs. - Start with 4–6 seconds while you tune the edit instructions. Increase duration after the look locks in. - Pick an auto aspect ratio when your inputs mix different source formats. Use fixed ratios when you need exact social framing. - Keep each reference focused on one job. Use video for motion, images for identity or style, and audio for rhythm. ## Limitations - Mini variants commonly cap output resolution at 720p. - Clip duration is short. Many hosts restrict outputs to about 4–15 seconds. - Reference inputs have hard limits. Video is usually capped at 3 clips, images at 9, and audio at 3. - Mixed reference sets may have a combined cap across all files. Some hosts limit the total to 12 items. - Low-quality inputs can break motion transfer. Heavy compression, shaky footage, fast cuts, and low light can cause drift and artifacts. - Real-person portrait references may be restricted. Some uses require identity verification or proof of legal authorization. ## Safety & compliance Follow the host platform’s content rules and acceptable use policy. Don’t generate illegal content. Don’t create non-consensual intimate imagery. Don’t attempt to generate or share child sexual abuse material. Get rights for any real person’s likeness, brand assets, or copyrighted characters. Some hosts scan and block harmful content and may report CSAM per legal requirements.