ByteDance / seedance 2.0 mini

## Overview Seedance 2.0 Mini is a short-form video generation model from ByteDance. It turns a written scene description into a finished MP4 clip. You can also guide motion using a starting frame, an optional ending frame, and extra reference images or audio. It’s useful when you need many fast variations that already include sound. ## What you can build - Product ads with camera moves like slow orbit, push-in, or macro pull focus - UGC-style vertical clips for short-form feeds - Storyboards and animatics from a single shot description - Before-and-after transitions by giving a first frame and a last frame - Style-locked series where multiple shots keep the same subject look using reference images - Music- or rhythm-led edits by conditioning motion on a short audio reference ## Inputs - A scene description written in natural language. Include camera direction, action, and mood. - An optional starting frame image for image-to-video. Provide a JPEG, PNG, or WebP file. - An optional ending frame image to steer the final look. Provide a JPEG, PNG, or WebP file. This requires a starting frame. - Optional reference images to keep a subject, wardrobe, product shape, or environment consistent. Provide 1–9 JPEG, PNG, or WebP images. In your prompt, refer to them as [Image 1], [Image 2], and so on. - Optional reference audio to influence timing and mood. Provide 1–3 MP3 or WAV files. Audio can’t be the only input. Pair it with a starting frame or at least 1 reference image. - Your target output resolution selection. This model tier supports 480p and 720p. - Your target aspect ratio selection. You can choose a fixed ratio like 16:9 or 9:16, or let it adapt. - Your target clip length selection in seconds. - A toggle to include synchronized audio in the generated video. - An optional prompt rewrite step that can remove brand names and reduce policy rejections. It keeps tokens like [Image 1] intact. - An optional watermark toggle for the generated video. - An optional numeric seed for repeatability. Set it to 0 to randomize. ## Outputs The model returns a generated MP4 video file. - Video content includes the motion, camera behavior, and scene changes described in your prompt. - The clip matches the chosen duration, resolution, and aspect ratio. - If audio generation is enabled, the MP4 includes a synchronized audio track with ambience, sound effects, and any prompt-directed dialogue. - The response can include a seed value so you can reproduce or vary a result. ## Limitations - Output resolution is limited to 480p or 720p in this Mini tier. - Clip length is limited to the supported short durations on this model. - Some input combinations may be restricted by the provider. For example, first/last-frame guidance may not be compatible with certain reference modes. - Face-heavy prompts or real-person likeness requests may be rejected by upstream filters. - Reference audio won’t work well if it’s noisy, clipped, or much longer than the allowed limit. - Low-quality inputs can degrade results. Blurry images, heavy compression, watermarked pictures, or inconsistent lighting often cause artifacts. - Text in the scene, logos, and small typography can drift frame to frame. ## Safety & compliance - Don’t use this model to generate non-consensual content, harassment, or sexual content involving minors. - Don’t create deepfakes of real people or impersonate public figures. - Don’t request copyrighted characters, brand assets, or licensed music you don’t have rights to use. - Expect automatic filtering. Some prompts, faces, and audio content can trigger rejections or forced edits. - Review outputs before publishing. Treat results as synthetic media and label them when required by your policy or local rules.