Dreamina generates controllable videos up to three minutes with timestamp controls
dreamina.capcut.com
Dreamina can now spit out AI videos up to three minutes long, and you can tell it exactly what happens at which timestamp. That's a huge jump from the usual 5-10 second AI clips everyone's used to.
Based on reporting by dreamina.capcut.com — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Most AI video tools still top out around ten seconds before things get expensive or the quality falls apart. Dreamina, ByteDance's video and image generation platform, just blew past that ceiling with a new mode that produces clips running up to three minutes, and it lets creators lock in specific actions or scene changes at chosen timestamps along the way.
That timestamp control is the real story here. Instead of typing one prompt and hoping the model holds together for the full runtime, users can essentially storyboard a video in segments, telling the system what should appear at the 20-second mark versus the 90-second mark. That's a meaningfully different workflow than the single-shot prompting most tools rely on, and it starts to look more like actual video editing than slot-machine generation.
Three minutes doesn't sound like much next to a feature film, but it's a big jump for a field where consistency across even 15 seconds has been a persistent headache. Characters drift, backgrounds warp, objects vanish. Stretching coherent output to three minutes while giving users checkpoints to steer the narrative suggests ByteDance has made real progress on temporal consistency, not just longer render times.
ByteDance has been aggressive in AI video, and Dreamina sits alongside tools like Jimeng in its stable of consumer-facing generative products. Longer, steerable video output pushes the platform closer to being a genuine short-form production tool rather than a novelty generator, which matters given how much of ByteDance's business, TikTok included, runs on short video content.
The obvious next question is quality at scale. Rendering three minutes with multiple directed beats is a much heavier lift than a single ten-second clip, and it remains to be seen whether motion stays smooth and characters stay recognizable across that whole runtime. But the direction is clear: AI video tools are racing toward longer, more controllable output, and timestamp-level direction might become the standard way people work with these systems going forward.
My take — AI-written commentary, not fact-checked reporting
Three minutes with timestamp control is a real step, not just a marketing number, and it's telling that a company sitting on the world's biggest short-video dataset is the one pushing this forward. Expect the
Read more about this at: dreamina.capcut.com