sora / text-to-video

OpenAI's Sora, as a ComfyUI node, with the whole generation running on Civitai's cloud instead of your GPU. You type a prompt, the job goes up through Civitai's Orchestration API, and a clip comes back as a native VIDEO socket ready to wire into a Save Video node. There's no OpenAI key involved and nothing to download - you pay in Buzz, like everything else in the official civitai-comfy-nodes pack.

The text-to-video variant is the purest form of the idea: prompt in, scene out. Its sibling, the image-to-video Sora node, is the one people usually graduate to once they have a still worth animating, but this node stands alone fine - especially for drafts where you're exploring what Sora does with an idea before committing a specific frame.

The inputs that matter

Only prompt is required. Everything else is optional and beginner-friendly:

• duration - 4 to 12 seconds, default 4. Start short: Sora is not the cheap engine in this pack, and seconds are the bill multiplier.
• resolution - 720p or 1080p, default 720p.
• aspect_ratio - auto, 16:9, or 9:16. With no image to anchor to, auto just picks whatever the model likes; set it explicitly if you need a specific framing for a feed or a screen.
• use_pro - the Sora Pro toggle, off by default. Better quality at a noticeably higher meter. Keep it off while you iterate; flip it on for the keeper.
• seed - set to lock a run, leave 0 for variety.
• api_config - auth socket.

Because there's no image input, the prompt carries the entire load. Sora understands camera language well, so write motion, shot, and pacing: "slow push-in on a street vendor's stall at dusk, steam rising, pedestrians passing in foreground." Static scene descriptions come back as boring clips; motion descriptions come back as Sora's signature show-off physics.

Outputs

The pack's standard trio: video (VIDEO) into your player/saver, workflow_id (STRING) to find the job in the Civitai sidebar, and raw_json (STRING) for the full response - the place to check what a run actually billed.

Install and auth

One-time, shared by every node in the pack. ComfyUI Manager → search Civitai Comfy Nodes → install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git
pip install -r civitai-comfy-nodes/requirements.txt   # just `requests`

Authenticate however suits your setup: wire a Civitai Auth node, set CIVITAI_API_TOKEN, or sign in from the Civitai sidebar. Headless/remote installs should use the env var.

Troubleshooting

• It's the expensive option. Sora sits at the premium end of this pack's video nodes, and OpenAI's model enforces its own content policy no matter who fronts the API. A refusal is the model saying no, not a broken install.
• Timeouts in the early days. The pack shipped June 2026 right as Civitai's GPU capacity was cut for crypto mining, and slow queues were the #1 real complaint - including users timing out on jobs. Raise timeout_minutes on the Civitai Auth node or set CIVITAI_COMFY_TIMEOUT if long jobs die.
• Output ignored your prompt? Check enable_prompt_enhancer-style behavior isn't the culprit - this node has no enhancer toggle, so what you typed is what Sora got. If results drift, it's Sora's own interpretation, and adding explicit motion/camera language is the fix.
• Early preview. The pack's README is upfront that nodes "may change without notice." If Sora's behavior shifts after an update, the generated nodes track Civitai's API spec - check the pack's changelog rather than blaming your graph.

Reach for this when you want Sora-quality motion and you'd rather pay per clip than for a machine that can't run it anyway. When you have a frame you love, the image-to-video sibling is the natural next node.