Canvas's "Generate video" options just grew by five. Seedance 2.0 Fast and Mini from ByteDance, Sora 2 Pro from OpenAI, and Veo 3.1 Fast and Lite from Google DeepMind are all live now, each backed by a real, callable model endpoint — not a placeholder.
What's new
Seedance 2.0 (Fast and Mini) — ByteDance's image-to-video model with synchronized audio, start/end-frame control, and motion prompts. Fast trades a little quality for lower latency and cost; Mini goes further in the same direction.
Sora 2 Pro — OpenAI's video model, generating dialogue lip-sync, environmental sound, and ambient audio directly from the same prompt that describes the scene.
Veo 3.1 (Fast and Lite) — Google DeepMind's model, animating a still image up to 4K with native audio. Lite is the lower-cost, lower-latency tier.
How to use them
Drag any of the five onto Canvas from the "Generate" section of the node palette, connect it to an image source the same way you would the existing Kling video node, and wire a prompt the same way as every other model here.
We've verified their endpoints and input schemas directly against each model's own documentation. As with any new integration, if a field behaves differently than expected, tell us — that's exactly the kind of thing we fix fast.