
Pass the ID of a completed image task as image_task_id in a new video request. Keep the image task and video task as separate records: one creates the visual reference, and the other creates motion from that reference. They share an account points wallet but have separate costs and completion states.
Key takeaways
- The source must be a successful image task owned by the same account.
- Send
image_task_idwithoutfile_id; the two reference selectors are mutually exclusive. - Budget for both tasks and keep a durable link between their IDs.
Approve the still before asking it to move
First inspect the completed image at its intended crop. A wrong face, illegible label or missing object will not become correct merely because the next request is a video. Select a still with enough room for the intended movement, and record what the animation should preserve.
The source task must report succeeded, with its result still available. A task that was accepted, is processing, or failed is not a usable reference. The API console documents this image-task reference path; the model hub supplies the current model options.
Link two tasks in your own data model
Store the dependency explicitly rather than replacing an image row with a video row. The following is an illustrative application record, not a payload for an Imagild endpoint:
{
"scene": "boat-passes-stone",
"imageTaskId": "11111111-1111-4111-8111-111111111111",
"videoTaskId": null,
"sourceApproved": true,
"motionBrief": "Track beside the boat as it passes the stone"
}
After submitting the video, populate videoTaskId from the new response's data.id. Keep both records for troubleshooting and cost reconciliation. If a user refreshes your page, recover these IDs from your own database instead of submitting the scene again.
Reference the image instead of uploading it again
The video body can look like this. Replace the illustrative UUID with the actual successful image task ID before submission:
{
"model": "MiniMax-H3",
"prompt": "The blue paper boat moves slowly past the stone. Keep the boat clearly visible while the camera tracks beside it.",
"image_task_id": "11111111-1111-4111-8111-111111111111",
"image_role": "first_frame",
"seconds": 4,
"resolution": "768P",
"aspect_ratio": "16:9",
"context_ir": false
}
Send this to POST /videos/generations with Bearer authentication, JSON content type and a new Idempotency-Key for this video operation. Do not send file_id as well. Reusing an image task avoids a client-side download-and-upload round trip, but it does not make the video generation free.
Calculate the combined cost once
A new 1K image currently uses 10 points. A four-second 768P video uses 60 points with enhancement off. Together, those two successful generations use 70 points; enabling enhancement on the video makes the total 75. An image that you already generated was charged for its original task; using its ID does not submit another image-generation request.
Membership eligibility remains separate from this arithmetic. Refer to the points and resolution guide when planning mixed image and video work. A retake is another generation, so keep it as another budget line.
Handle dependency failure before making a new request
If the reference belongs to another account, is not an image task, has not completed or no longer has a usable result, stop the dependent operation and correct the reference. Do not keep changing the video prompt while the real issue is the source ID. The reference attachment audit offers a useful broader check: verify the actual bound asset, not merely its displayed label.
Poll the new video task independently. The distinction between accepted work and finished work also appears in HTTP 202 semantics. Once it succeeds, download and review the full clip against the approved still and motion brief. Check preserved details at several points, because a convincing opening frame does not prove identity or object continuity throughout the shot.
Related tasks
- Imagild Points, Resolution and Jobs: Plan a Creation Before You Spend
- After Reordering References, Does “Image 1” Still Point to the Right File?
Sources and scope
- Imagild API integration guide: Actual asynchronous request fields, reference selectors, access conditions, retries and result download.
- Imagild model hub: Current H3 model ID, duration, resolution, ratios and displayed points.
- HTTP 202 Accepted: General meaning of acceptance versus eventual completion; not a guarantee of Imagild task success.
Product instructions were checked against the current Imagild API contract and the public model catalog. Code, budgets and acceptance checks are illustrative; this article does not claim a new paid generation test, an uptime measurement, a search-volume study or a ranking of AI citations.
Prepared by Imagild Editorial, the company editorial team responsible for these guides. AI assisted research and drafting. This is documentation-based guidance with proposed examples, not a claim of hands-on testing or an independent product review. Report a correction.
Cover: Original Imagild product screenshot captured on 28 October 2026; model availability and prices may change. This image is not an API response or a performance benchmark.