Generated facts come from
packages/schema/src/v2/models.ts. Edit numbers there, not here.
OpenAI's video model via EvoLink, with native audio, text-to-video only.
| Kind | video |
| Tier | premium |
| Status | candidate |
| Provider | evolink |
| Modes | text-to-video, image-to-video |
| Features | native-audio |
| Limits | 4 to 12 s (per mode below); worker waits up to 30 min |
| User price | none (candidate) |
| Quality / speed | premium / slow |
| Verified | no recorded run yet |
The mode is derived from the request: first_frame means image mode, refs / video_refs / audio_refs mean reference mode, neither means text mode.
| Mode | Provider model | Inputs | Seconds | Aspect | Quality | Seed | Verified |
|---|---|---|---|---|---|---|---|
| text | sora-2-preview |
prompt only | 4 to 12 | 16:9, 9:16 (default 9:16) | 720p | no | no recorded run yet |
| image | sora-2-preview |
first_frame | 4 to 12 | 16:9, 9:16 (default 9:16) | 720p | no | no recorded run yet |
- text: Only 4, 8 or 12 s. Strict content moderation; real people are not accepted.
- image: One image; its pixel size must match the aspect exactly (1280x720 or 720x1280). No real people.
- No reference input, so a saved character's identity cannot be locked. Planned for cinematic b-roll only.
- Not selectable yet.