FL Fal Sora 2
OpenAI's Sora, text- or image-to-video, with native audio
- image
- frames
- audio
- video_url
- status_msg
OpenAI's Sora 2, both text-to-video and image-to-video, through Fal.ai. Sora is one of the two closed video models the wider AI-art community explicitly measures open-source video against - Google's Veo 3 is the other - and it's still ahead of local alternatives on raw quality, even as the gap keeps closing generation over generation. This is the node if you want a Sora shot inside a ComfyUI pipeline instead of the ChatGPT/Sora app.
The mechanism
Text-to-video when image is left empty, image-to-video when you supply a starting frame. use_pro switches which Sora tier gets hit - Fal exposes both a standard and PRO variant, and the node picks the right endpoint automatically based on this toggle rather than you choosing it directly. openai_api_key is a separate, optional pass-through field distinct from fal_api_key, which is worth knowing: this node can involve two different providers' keys, with Fal hosting the call and optionally forwarding your own OpenAI credentials through it.
Inputs and outputs that matter
fal_api_key (required) is your Fal.ai key - the one that actually authenticates the call. prompt defaults to a placeholder, "A dramatic scene..." - replace it, it's not worth keeping. use_pro toggles the PRO tier. resolution is auto/720p/1080p. aspect_ratio is auto/16:9/9:16 - note that "auto" is listed as an available choice even though the field's own default is "16:9". duration is 4, 8, or 12 seconds, an enum of actual integers rather than strings the way some of this pack's other duration fields work. openai_api_key is optional and easy to leave blank without anything breaking. image (optional) enables image-to-video. nth_frame (1-4) is the frame-extraction stride on the result.
Outputs: frames, audio (Sora generates audio natively alongside video - one of the two things, along with Veo 3, that no open-source model currently does at all), video_url, and status_msg.
Installing it
ComfyUI Manager: search Fill-Nodes, install, restart. Manually: cd ComfyUI/custom_nodes && git clone https://github.com/filliptm/ComfyUI_Fill-Nodes, then restart.
Where people get burned
This is the most expensive node in the whole pack to run casually. Sora access through any API is priced accordingly, and use_pro on top of 1080p/12s duration is a genuinely costly single call - test at the shortest duration and lowest resolution first, and only scale up once your prompt is doing what you want.
Double-check which key goes where, too: fal_api_key is required and does the actual authenticating, openai_api_key is optional. Don't assume you need an OpenAI account on top of a Fal one unless the specific behavior you're after calls for it.
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| fal_api_key | STRING | — | |
| prompt | STRING | A dramatic scene... | — |
| use_pro | BOOLEAN | false | — |
| resolution | COMBO | auto | 3 options: auto, 720p, 1080p |
| aspect_ratio | COMBO | 16:9 | 3 options: auto, 16:9, 9:16 |
| duration | COMBO | 4 | 3 options: 4, 8, 12 |
| openai_api_key | STRING | — | |
| nth_frame | INT | 11–4 | — |
| imageopt | IMAGE | — |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| frames | IMAGE | — |
| audio | AUDIO | — |
| video_url | STRING | — |
| status_msg | STRING | — |