Create
ByteDance Just Buried Text Prompts: New AI Makes Films Using Other Videos, Not Words

ByteDance Just Buried Text Prompts: New AI Makes Films Using Other Videos, Not Words

Arkadiy Andrienko

Chinese tech giant ByteDance, the parent company of TikTok, has officially launched the multimodal version of its AI video synthesis model—Seedance 2.0. Unlike its predecessors and most Western counterparts, this model doesn’t just take text prompts or static images. It accepts ready-made video clips, audio tracks, and any combination of up to 12 files in a single request.

The real game-changer here isn’t just visual quality—it’s the shift in how you control the output. Before, if you wanted a specific camera move or a character’s facial expression, you had to describe it in painstaking detail, typing out paragraphs in a prompt. Now? Just upload a reference—say, a 5-second clip with the exact shooting style you’re after—and write: “@video1, replicate the cinematography.” The model doesn’t just borrow composition; it picks up on editing rhythm, gaze direction, and even syncs visuals to the beat or tone of your audio.

Seedance 2.0 supports up to 9 images, 3 videos, and 3 audio files per session. Each generated clip maxes out at 15 seconds, with resolution up to 2K. But its standout trick—one that’s been notoriously tough to pull off—is reference stitching. You can ask the AI not just to generate a new segment, but to insert it between two existing clips while perfectly matching the style, lighting, and even background noise.

That said, tighter safety policies mean the platform has nixed generation involving realistic human faces. The system automatically blocks uploads of photos or videos where living individuals are clearly identifiable. Same goes for face-swapping in clips.

ByteDance Just Buried Text Prompts: New AI Makes Films Using Other Videos, Not Words

For now, the model is only accessible through the Chinese version of Dreamina, which is tied to a Douyin account. International apps (CapCut and Imagine.Art) are expected to get the integration by the end of February.

So—does Seedance 2.0 actually have what it takes to replace video editors in pre-production, or is it still just a tool for spitballing rough drafts? Drop your thoughts in the comments.

    About the author
    Comments0