POST /api/v1/videos/generations with model: gemini-omni-1-2-flash to get a 3-15 second clip at 24fps with a native audio track. Start from text, a first frame (image_url), a first and last frame, or up to 10 reference images in image_urls. Output runs from 360p to 4K; track the job with the returned img_uuid.
prompt. There is no switch, so turning sound / generate_audio off is rejected.
image_url to start from your image, and add image_end_url to interpolate to a final frame. With a last frame, duration can be at most 10 seconds.
image_urls guide the video without being used as exact frames. References cannot be combined with image_url / image_end_url.
gemini-omni-1-2-flash.image_url. Cannot be combined with image_urls.duration: integer 3-15, default 10 (at most 10 with a last frame). resolution: 360p, 720p (default), 1080p or 4k; 1080p and 4k are upscaled. ratio: 16:9 (default) or 9:16.negative_prompt, seed, middle keyframes, mixing frames with image_urls, and video input are rejected before any tokens are charged.GET /api/v1/jobs/detail when the task status changes, with up to 3 retries.model, prompt and any frames or reference images to /api/v1/videos/generations.img_uuid and poll GET /api/v1/jobs/detail, or pass custom_callback_url.success, download the video from the result and store it on your side.image_urls with a spoken line in the prompt for short ads that already have a voice-over.image_url, with ambient sound included.9:16 clips at 360p for drafts and re-run the chosen prompt at 1080p or 4k.No. Every clip has a native audio track and requests that set sound or generate_audio to off are rejected.
Yes. prompt is required in every mode, up to 3000 characters.
No. Use either image_url / image_end_url or image_urls in one request.
3 to 15 seconds, default 10. With image_end_url the limit is 10 seconds.
No. They are upscaled by the model; 2160p is accepted as 4k.
Not yet. Video input is rejected before any tokens are charged.