Seedance 2.5 Reference Uncensored

Overview

Seedance 2.5 Reference Uncensored is a ByteDance video model tuned for reference-guided generation. You guide a new clip with a written prompt plus optional image, video, and audio references. The model uses those references to lock identity, style, motion, and timing across a single continuous take. It outputs a finished video file, and it can include a synchronized audio track when you enable it.

What you can build

• Product ads that keep the same packaging, label text, and lighting across a full shot
• Character-consistent scenes using a small set of reference images
• Motion transfer by borrowing camera movement and pacing from a reference clip
• Style transfer that keeps motion from video while changing wardrobe, setting, or art direction
• Music- or beat-driven visuals by syncing cuts and motion to a reference audio track
• Video edits where you restyle an existing clip while keeping its length and framing
• Video extensions that continue a reference clip with new footage that matches it

Inputs

• A video prompt that describes the subject, action, setting, camera language, and (optionally) audio cues. When you add references, say what to keep and what to change.
• Optional reference images (JPEG), used to hold faces, outfits, props, branding, or overall visual style. Provide 1 to 30 images. In the prompt, refer to them as “[Image 1]”, “[Image 2]”, and so on.
• Optional reference videos (MP4 or MOV), used to borrow motion, pacing, shot rhythm, or to edit or extend an existing clip. Provide 1 to 10 videos. Each clip must be about 2 to 30 seconds, and the combined total must stay within about 30 seconds.
• Optional reference audio (WAV or MP3), used to guide rhythm, timing, ambience, or voice style. Provide 1 to 10 audio clips. Each clip must be about 2 to 30 seconds, and the combined total must stay within about 30 seconds.
• An output resolution choice. This model supports 480p or 720p output.
• An output aspect ratio choice. You can pick a fixed ratio (like 16:9, 9:16, or 1:1) or let the model pick automatically. If you provide a reference video, the output keeps that video’s aspect ratio.
• A target duration in seconds, or automatic length selection. If you are editing a reference video, automatic duration is required so the output keeps the input clip’s length.
• An option to generate synchronized audio with the visuals. Turn it off if you need a silent clip.
• An output container format choice (MP4 or MOV). MP4 is the safest for playback. MOV is often preferred for post-production workflows.
• An option to add a watermark to the generated output.

Outputs

• One rendered video file in MP4 or MOV format.
• Video resolution matches your selection (480p or 720p).
• Duration matches your selection, unless the request is interpreted as an edit of a reference video, in which case the output keeps the input video length.
• If audio generation is enabled, the file includes an audio track aligned to the visuals. This may include ambience, sound effects, music, and spoken dialogue when your prompt includes it.

Recommended settings

• For iterating on prompts, generate at 480p first. Then re-run at 720p for the final.
• If you provide a reference video and want a true edit, set duration to automatic. Let the output inherit the reference video’s length and aspect ratio.
• When you want a specific framing (vertical, square, ultrawide), set a fixed aspect ratio and avoid reference videos that conflict with it.
• Use MP4 for general sharing. Use MOV when you plan to color grade, key, or composite.

Limitations

• Reference limits apply. You can attach up to 30 images, 10 videos, and 10 audio clips, but reference video and audio are capped by total combined duration.
• Reference videos must be short. Each clip must be about 2 to 30 seconds, with about 30 seconds total across all reference videos.
• Reference audio must be short. Each clip must be about 2 to 30 seconds, with about 30 seconds total across all reference audio.
• Some inputs may be rejected. Providers may refuse prompts or references that contain real human faces, or other sensitive content.
• Low-quality inputs reduce control. Blurry images, heavy compression, inconsistent lighting, fast cuts, shaky camera, or noisy audio can cause drift and artifacts.
• Ambiguous references can confuse ownership. If multiple people look similar across references, identity can swap mid-shot.

Safety & compliance

• “Uncensored” does not mean “no rules.” The provider and platform can still block content and references.
• Only use reference media you own or have rights to use. Don’t upload copyrighted clips, logos, or recognizable protected characters unless you have permission.
• Don’t create sexual content involving anyone under 18, or content that depicts non-consensual acts. Avoid content that exploits or sexualizes youth.
• Don’t upload real-person likenesses without consent. Even with consent, face references can be refused by the provider.
• If you need a mature theme, describe it clearly but keep it lawful and non-exploitative. Expect moderation variance across prompts and inputs.