OpenAI Sora - Video (DEPRECATED)

Here's the thing you need to know first: this node is on a countdown. OpenAI is shutting down the Sora v2 API in September 2026, and when that happens this node gets removed from ComfyUI. It says so right in the node's own description. So if you're here to build a durable, long-term workflow, close this tab and pick something else - this page is for the people who want to use Sora while it's still running.

When it works, it's exactly what the "Sora" name promised back in 2024: OpenAI's video model, reached from inside your ComfyUI graph. Like every node in the built-in partner/video category, it's an API node. It ships with ComfyUI core, uploads your optional image to Comfy Org's backend, calls OpenAI's video endpoint through Comfy's proxy, polls until the clip is done, and downloads it back as a VIDEO. No local model, no key to OpenAI directly, billed to your Comfy account in credits.

The inputs that matter

• prompt - guiding text. The tooltip notes it can be empty if you're providing an input image, which is handy for pure image-to-video.
• image - an optional reference image. One image only; the node rejects batches with a clear error.
• model - sora-2 or sora-2-pro. The pro tier runs roughly double the per-second price and unlocks the bigger frame sizes.
• size - 720x1280, 1280x720 for the base model; sora-2-pro also allows the tall 1024x1792 and 1792x1024 formats. Note the base model rejects anything outside its two landscape/portrait options.
• duration - 4, 8, or 12 seconds. The price badge scales per second.
• seed - cosmetic, as usual for this family: it only decides whether the node re-runs, and results are nondeterministic either way.

Output is a single VIDEO object.

The honest take

The community's relationship with Sora is nostalgia. The "Nope, you can't have it!" era - when Sora demos were locked behind an API and all local users had was AnimateDiff - is the reference point the open ecosystem still measures against, and by 2026 the running joke is that local Wan clips look better than those original Sora teasers. That doesn't mean Sora 2 is bad; it means you should be honest about why you're using it. If you want a specific OpenAI quality bar without a separate Sora subscription, this node gives it to you inside ComfyUI. If you want a workflow you can still run next year, this is the wrong node by construction.

Where people get burned: the deprecation itself, mostly. Also the usual API-node taxes - paid credits, a queue that can take minutes, and no deterministic output. And watch the model/size combo: requesting 1024x1792 on the base sora-2 model fails validation, because that format is pro-only.

Use it while it's here if you must, and don't build a production pipeline on it. In a few months this page will be a footnote about the six months Sora lived inside ComfyUI.