Seedance 2.5

About the model

Seedance 2.5 is the next generation of ByteDance’s video model. The main difference from Seedance 2.0 is length: the model generates a single continuous clip of up to 30 seconds, so one request fits a complete scene with a setup and a payoff rather than a single shot.

Its second strength is reference handling. The model reads attached images and audio more accurately: it picks up composition and visual language, and keeps characters and products looking the same from the first second to the last. That makes it a good fit for branded content and ads, where recognizability matters.

For a fast, low-cost option to test ideas, see Seedance 2.0 Mini. For short clips up to 15 seconds, Seedance 2.0 is still available.

Strengths

• Clip lengths from 4 to 30 seconds — twice as long as Seedance 2.0
• Precise understanding of references — images and audio files
• Keeps characters, products, and style consistent across the whole scene
• First and last frame mode for controlled motion
• Native synced audio generation
• 480p, 720p, and 1080p output, aspect ratios from 21:9 to 9:16
• Web search for enriching prompt context

Best for

• Product launch videos and ad assets
• Short films, trailers, and story scenes
• Branded content with consistent style and characters
• Ecommerce and marketplace videos

Limitations

• Maximum length is 30 seconds
• Up to 30 images and 10 audio files per request; total audio length up to 30 seconds
• Image size up to 30 MB, audio files up to 15 MB
• Built-in adult content filter

Prompting tips

• Describe the scene step by step — what happens at the start, middle, and end of the clip
• For longer clips, set the pacing and shot changes right in the prompt
• Attach images of a character or product from several angles so the model keeps its look
• Use either first and last frames or references — they don't work together
• Use 480p and a short length for tests, 1080p for the final cut