New in AI Studio: Seedance 2.5 Brings 30-Second Multimodal Video Generation to Prizmad

Prizmad's AI Studio added another video model this week: Seedance 2.5, ByteDance's latest generation engine, live now at prizmad.com/tools/seedance-2-5. It generates coherent AI video up to 30 seconds long from a text prompt, a still image, or up to 50 combined image/video/audio references — with native synchronized audio generated in the same pass as the visuals, not layered on afterward.

What it actually does

Seedance 2.5 runs in three modes, each suited to a different starting point:

• Text-to-Video — describe the subject, scene, action, camera movement, lighting, and sound; the model generates the clip from scratch.
• Image-to-Video — upload a start image (and optionally an end image) and the model animates it into motion, following the input image's own aspect ratio automatically.
• Reference-to-Video — combine up to 30 reference images, 10 reference videos, and 10 reference audio files (50 total, capped at roughly 30 seconds of combined reference video/audio) into one generation, addressing individual files in the prompt as @Image1, @Video1, and @Audio1.

Output runs 4 to 30 whole seconds at 480p or 720p, across seven aspect ratios (Auto, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16). Audio — ambience, effects, and dialogue — generates alongside the visuals by default, and per the tool's own interface, at no additional token cost beyond the video generation itself.

What it costs

The AI Studio catalog lists Seedance 2.5's default generation — a 10-second, 720p, text-to-video clip — at 10 tokens. Cost scales with output duration, resolution, generation mode, and whether video references are used (reference-to-video generations with video inputs cost more per second than text- or image-to-video). The tool is asynchronous — you submit a job and it processes in the background — with an estimated turnaround of roughly 5–14 minutes depending on queue load and the settings chosen.

Why this matters for ad creative specifically

Seedance 2.5's most relevant capability for ad production is its native 30-second single-pass generation — earlier-generation models typically needed multiple shorter clips stitched together, which shows as visible seams or subject drift over longer runtimes. A single continuous 30-second generation keeps a product or presenter's appearance consistent across the whole clip, which matters for anything longer than a quick hook.

The reference-to-video mode is the more experimental option: feeding in existing brand footage, a product reference video, and a specific audio track in one generation is a different workflow than a typical prompt-only text-to-video tool, and it's worth testing directly against your own source material rather than assuming behavior from the spec sheet alone.

Try it

Seedance 2.5 is live now inside the same AI Studio that ships 30+ tools on Prizmad's standard subscription — no separate signup or API key needed. At 10 tokens for a default generation, it's a reasonable first test against your own product shots or brand footage before committing to a longer or higher-resolution run.