Introduction
Gemini Omni 1.2 Flash reference-to-video builds a new scene around uploaded visual guidance. Provide a prompt and at least one of reference_image_urls or reference_video_urls. Images guide people, objects or places; short clips guide look and motion. The reference soundtrack is ignored.
Google says 1.2 improves consistency across up to six reference images and recommends up to three pictures per character for up to two characters. Video references can include up to three clips, each up to three seconds. For an 11–15 second result, Google lists one reference clip; in our tests, an 11-second output with two clips also completed. Use the listed guidance when planning a long scene.
Key features & highlights
- Combine image references and video references within the same generation request.
- Assign visual roles to as many as six uploaded images.
- Guide motion or look with up to three reference clips, subject to Google’s longer-output guidance.
- Name each upload in ordinary wording, using Image 1 or Video 1 in upload order.
- Generate a new audio track with the scene; audio from reference footage is ignored.
Request parameters
| Parameter | Required | Description |
|---|---|---|
prompt | Yes | Required. Explain what to use from each numbered image or video, then describe the new action and sound. at least 1 characters · at most 20000 characters |
reference_image_urls | No | Up to 6 image URLs. Supply images, video references, or both. at least 1 items · at most 6 items |
reference_video_urls | No | Up to 3 video URLs, each clip up to 3 seconds. These guide look and motion; their audio is ignored. at least 1 items · at most 3 items |
aspect_ratio | No | The default is landscape 16:9; vertical 9:16 is also available. Framing does not add a charge. |
resolution | No | Select direct 360p or 720p (default), or choose 1080p/4K. The 1080p and 4K options are upscaled by Google and cost more. |
duration_seconds | No | Choose 3–15 whole seconds; default 8. For 11–15 seconds, follow Google’s one-video-reference guidance. |
Pricing
Reference generation bills the requested output seconds using the selected resolution rate. Image count, reference clips and aspect ratio do not add charges. You pay nothing when a reference request fails or Google refuses it.
| Specification | Price |
|---|---|
| $0.0276 / second |
| $0.084 / second |
| $0.126 / second |
| $0.252 / second |
Use cases
- Place two referenced travelers in a railway dining car for a short exchange.
- Carry a product from its reference photo into a newly described location.
- Use a short motion reference to guide the movement of a new market scene.
Prompt guide & pro tips
- Assign each person a clear source, such as “the traveler in Image 1,” using upload order.
- Explain what each video guides, such as “move like Video 1,” rather than assuming its role.
- Describe interaction separately from appearance, such as “slide a folded note across the table,” to define the new event.
- Follow Google’s character guidance with a phrase such as “use Images 1–3 for the first character,” while staying within six images total.
- Describe fresh sound, such as “steady train rhythm and clinking cutlery,” since audio in the reference clips is ignored.
API quick start
# Generate the idempotency key once; reuse it when retrying after a lost response.
export SPICY_IDEMPOTENCY_KEY="${SPICY_IDEMPOTENCY_KEY:-$(uuidgen)}"
curl -X POST https://api.spicyapi.ai/api/v1/jobs/createTask \
-H "Authorization: Bearer $SPICY_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: $SPICY_IDEMPOTENCY_KEY" \
-d '{
"model": "google/gemini-omni-1.2-flash/reference-to-video",
"input": {
"aspect_ratio": "16:9",
"duration_seconds": 8,
"prompt": "A slow aerial shot drifting over a misty pine forest at sunrise",
"resolution": "720p"
}
}'
# Copy data.taskId from the response into TASK_ID, then poll the status.
TASK_ID="TASK_ID_FROM_RESPONSE"
curl "https://api.spicyapi.ai/api/v1/jobs/recordInfo?taskId=$TASK_ID" \
-H "Authorization: Bearer $SPICY_API_KEY"
See the API reference for authentication, the complete schema, polling, and error handling.
