ByteDance / seedance 2.5

Overview

Seedance 2.5 is a generative video model from ByteDance. You describe a scene in text, then it synthesizes a new video clip that matches your direction. You can also anchor the motion with a starting frame, and optionally force an ending frame. This helps you turn an idea or still into a shot you can edit, cut, and publish.

What you can build

• Short ads and product shots with controlled camera moves
• Vertical social clips for reels, shorts, and stories
• POV action sequences, sports moments, and travel b-roll
• Animated posters and key art by turning a still into motion
• Before-and-after style reveal clips with a guided last frame
• Reference-driven clips that keep a character, an outfit, or a product consistent through the shot
• Atmosphere clips with synchronized sound design and music

Inputs

• A scene description written in natural language. Include subjects, actions, camera movement, and style.
• An optional first-frame image (JPG or PNG recommended) to start from for image-to-video. The output keeps this image’s aspect ratio.
• An optional last-frame image to steer how the clip ends. This only works if you also provide a first frame.
• An optional set of reference images, up to 30, that lock in a character’s face, an outfit, a product, or a visual style. Call them out in your description as [Image 1], [Image 2], and so on. Reference images replace first and last frame guidance rather than adding to it: when you supply them, the model composes the opening shot itself instead of starting from your image.
• A target resolution choice for the generated clip. Pick 480p for lighter outputs or 720p for sharper detail.
• A clip duration you choose in seconds, from 4 to 30.
• An option to include synchronized audio. Turn it off if you need silent footage.
• A video container choice: MP4 for maximum compatibility, or MOV when you want higher color precision.
• An optional prompt rewrite step that removes brands and risky content cues before generation.
• An optional watermark toggle.

Outputs

The model returns a single generated video file.

• File format: MP4 or MOV
• Length: the duration you requested, between 4 and 30 seconds
• Resolution: 480p or 720p
• Aspect ratio: either chosen by you, chosen automatically, or locked to your input frame
• Audio track: included only when you enable synced audio
• Watermark: included only when you enable watermarking

Recommended settings

• Start with 480p and 4–6 seconds while you iterate on the prompt.
• Switch to 720p once the motion and composition look right.
• Turn on prompt rewriting when you enable synced audio.
• Use MP4 for posting and reviews.
• Use MOV when you plan color grading, keying, or compositing.
• Provide a first frame when you must keep a specific framing and aspect ratio.
• Add a last frame when you need a clear end pose or end composition.
• Reach for reference images instead of a first frame when the subject matters more than the exact opening framing. Three to nine clean, consistent images of the same subject work best, and referring to them by number in the prompt keeps them straight.

Limitations

• The maximum clip length is 30 seconds per generation.
• Output resolution is limited to 480p or 720p in this configuration.
• If you provide a first frame image, the output keeps that image’s aspect ratio.
• A last frame guide requires a first frame guide.
• Reference images cannot be combined with a first or last frame in the same generation.
• The provider may reject requests that include human faces.
• Motion can drift in long clips, especially with fast camera moves.
• Heavy compression, blurry images, and low-light inputs reduce detail and increase artifacts.
• Fine text, logos, and small UI elements often deform or flicker across frames.

Safety & compliance

• Expect content filters. Some prompts and images will be blocked.
• Don’t request real-person impersonation or non-consensual likeness content.
• Avoid brand names, copyrighted characters, and copyrighted music cues in prompts.
• If you enable synced audio, keep prompts free of trademarked lyrics or recognizable artist references.
• Only upload images you have rights to use and share.