Through through the API, sora-2 can only generate 4, 8, or 12 second videos. They “cut off” unexpectedly, like mid-sentence the video will just end. If there is a solution for that, I would love to hear it.
I tried to switch to sora-2-pro, but it doesn’t seem to allow longer videos, I get constrained to the same 4, 8, or 12 seconds, attempting the 10, 15, or 25 second videos just gets rejected. Am I missing something?
You seem to be reading into something beyond the API reference. You missed stopping there.
POST https://api.openai.com/v1/videos
Parameters:
prompt: string # required
input_reference: file # optional
model: string # optional (default: sora-2) [sora-2, sora-2-pro]
seconds: string # optional (default: 4) [4, 8, 12]
size: string # optional (default: 720x1280) [720x1280, 1280x720, 1024x1792, 1792x1024]
I see. So the API limits a max of 12 seconds (even for pro), even though longer videos can be generated outside the API.
This is unfortunate.
What’s up with sharp cutoff on videos though? Is that normal? It’s just midway through a sentence and the video just ends.
I suspect the length of the output, and fitting content within, is just something you’ll need to prompt well - without being given a prompting guide by OpenAI.
Generative AI produces sequences, and like “chat”, might not know when you have forced a maximum output cutoff point, and the model might have training on longer outputs than can be delivered by API.
The AI making noise is mostly a side-effect, rarely useful except for novelty. A speaking video might get the mouth moving for you, for an ADR voice actor to make different generations consistent.
Prompting guide about “timing”.
That’s helpful, thanks. Sounds like I’m going to have to go in the direction of generating video and stitching it with audio.