AI video generators push past 15 seconds

AI video tools that once stopped at single short clips are starting to ship longer, stitched workflows. Topview’s site still labels Seedance 2.0 generations at 15 seconds, while Topview’s March 24 video says its Agent V2 can turn those clips into “multi-scene long-form video creation.” (topview.ai) (youtube.com) AI video generation means software makes moving images from text, pictures, audio, or reference footage instead of a camera shoot. ByteDance’s Seedance 2.0 says it accepts text, image, audio, and video inputs so users can control motion, lighting, camera movement, and rhythm from uploaded references. (seed.bytedance.com) (seedance2.app) The public Seedance 2.0 web app still caps output at 720p and 15 seconds, according to its documentation. That same documentation says users can upload up to 9 images, 3 videos totaling 2 to 15 seconds, and 3 audio files up to 15 seconds each. (seedance2.app) Topview’s own homepage shows “Seedance 2.0 16:9 480p 15s,” and its March 24 YouTube description says Agent V2 “breaks the limitation” of 15-second clips by generating scenes sequentially on a single timeline. The pitch is not one longer raw generation from the underlying model; it is a workflow that plans, generates, expands, and splices multiple scenes inside one editor. (topview.ai) (youtube.com) That shift puts the bottleneck in a different place. Instead of asking one model for a full minute of coherent action, these products ask for several shorter shots and then handle storyboard, continuity, and assembly in the same interface. (youtube.com) (seedance2.app) The 15-second number matters because many rival tools still live in that range. Runway says Gen-4 creates 5- or 10-second videos, while Kling 3.0 says it can generate up to 15 seconds in a single pass. (help.runwayml.com) (klingapi.com) Google is pushing on a different axis: model capacity rather than consumer workflow. Google DeepMind says Veo is its state-of-the-art video model, and Google Cloud’s Veo 2 documentation describes it as the stable video line on Vertex Artificial Intelligence, though consumer-facing tools have often exposed shorter clips than the model’s headline capabilities. (deepmind.google) (docs.cloud.google.com) The recent creator videos around Seedance 2.0 focus less on raw model research than on practical access: walkthroughs, prompt structure, and how to stretch free credits before paying. That matches the market Topview targets on its site, where it sells an “AI Video Agent” for marketing, product ads, and social clips rather than a research demo. (youtube.com 1) (youtube.com 2) (topview.ai) ByteDance’s Seedance page frames the model as a tool for “industry standards” and multimodal control, while Topview frames it as one component inside a broader ad-making stack. The result is that “longer video” now often means longer finished output assembled from many short generations, not the end of short clip limits inside the base model itself. (seed.bytedance.com) (topview.ai) For creators, the immediate change is simple: the ceiling is moving from clip length to workflow length. The next test is whether these stitched timelines can keep characters, motion, and style consistent enough that viewers stop noticing where one 15-second generation ends and the next begins. (youtube.com) (seed.bytedance.com)