☁️BizyAir Sora2 Text To Video

Sora spent its first year as the video model everyone talked about and almost nobody could actually run - the community's own framing at the time was "Nope, you can't have it," while local alternatives like AnimateDiff looked primitive by comparison. Open video models closed a lot of that gap since, but Sora 2 is still closed, still API-only, and this node is BizyAir's way of reaching it without you setting up a separate OpenAI account or API integration.

What it does

It's deliberately the simplest node in this pack's video lineup - just prompt and model are required. model picks between sora-2 and the sora-2-pro variant; the naming convention strongly implies pro trades higher cost for higher quality, in line with how OpenAI tiers its other products, though the exact difference in output isn't exposed in the node itself.

Everything else is optional and coarse by design: aspect_ratio is just 9:16 or 16:9 (no square, no ultra-wide), duration is either 10 or 15 seconds flat, and size is small or large rather than an explicit pixel resolution the way BizyAir's Doubao/Seedance nodes expose. If you're used to Seedance's granular resolution-plus-ratio controls, this node feels sparse by comparison - that's Sora's API surface, not a limitation BizyAir added.

Output

A video, plus the bizyair_model_name passthrough label confirming which of the two model tiers actually ran.

Installing it

ComfyUI Manager: search BizyAir, install, restart. Or by hand:

cd ComfyUI/custom_nodes && git clone https://github.com/siliconflow/BizyAir.git

Restart after. Nothing downloads locally - this is a thin client into OpenAI's hosted Sora 2 via BizyAir's proxy and billing layer. A BizyAir account with an API key set is required before this node produces anything; the pack walks you through setup on first use.

Where people get burned

Expecting fine-grained control you don't get here. No seed, no camera controls, no negative prompt, no resolution number - just prompt, model, aspect ratio, duration, and size. If your workflow depends on reproducibility or precise framing, this node isn't built for that; it's built for "describe it, get a video."

Underestimating cost on sora-2-pro. Closed frontier video models are the most expensive category of BizyAir credit spend across this whole pack - test a prompt on the base sora-2 model before committing to the pro tier for a batch.

Assuming size means resolution in the usual sense. small/large is a coarser knob than the literal pixel dimensions other BizyAir video nodes expose - don't expect to dial in an exact frame size here.

Forgetting this is a different vendor from ByteDance's Seedance nodes elsewhere in this pack. They live in the same node pack and look superficially similar, but Sora is OpenAI's model with OpenAI's own content policy and quirks - don't assume behavior you learned on Seedance carries over.