We just released Seedance 2.5, ByteDance's newest video model, and I spent a while making sure the full parameter surface is exposed as plain form controls rather than a dumbed-down "magic prompt box". Sharing what it does and where it falls short.
What a single generation gives you: 4 to 30 seconds of video, you pick the exact second. 480p or 720p, MP4 or MOV. Audio comes out in the same pass, meaning speech, sound effects and background music. The convention that matters: wrap spoken lines in double quotes in your prompt and they become dialogue, everything outside quotes reads as scene direction.
Input modes beyond plain text: image-to-video with a first frame and optional last frame, or a reference mode that takes up to 30 images + 10 videos + 10 audio tracks in one request. A lone audio file is a valid input, so you can start from a finished song and generate a video to fit it. Aspect ratios run from 21:9 to 9:16, with an adaptive default if you would rather let the model decide.
The small feature I ended up liking most: the generator can hand back the closing frame as a still image. Feed that into the next run as the first frame and you can chain past 30 seconds with pixel continuity instead of re-describing the previous shot.
Honest limits so nobody is surprised: 720p is the ceiling, so 4K delivery means an upscale step. No seed parameter, no reproducible runs. Generations cost credits scaled by length and resolution, so the sane workflow is drafting at 5 seconds 480p and only committing to 30 seconds once the prompt is settled. And to be clear about positioning: the model belongs to ByteDance, we are a third-party front end for it.
Page here if you want to look at the parameter surface:
https://seedance2.so/seedance-2-5
Would love to hear what breaks or what control you would want that is missing. That feedback directly turns into config changes.
Select Language