
Enable H3 enhancement when you have a specific shot to evaluate and are willing to pay five additional points for that video task. The API switch is context_ir, a JSON Boolean that defaults to false. It exposes an optional generation setting; it does not guarantee better motion, accurate lettering or seamless continuity.
Key takeaways
- Use
context_ir: trueto request enhancement, not the string"true". - The surcharge is five points per video task, added to the duration and resolution charge.
- Compare finished clips against written acceptance criteria before adopting enhancement as your default.
Identify the problem before changing the switch
Write a short statement such as “the hand must remain around the cup through the final second.” That is more useful than asking whether one clip looks generally more impressive. If the prompt is ambiguous or the source image shows the wrong object, correct those inputs first.
The Imagild API guide describes the supported request, while the model hub shows the current options and point costs. Enhancement is one parameter within that contract. It does not enable a new endpoint, introduce reference-video support or remove the 4–15-second duration limit.
Approve the comparison budget
The following figures cover one video generation at 768P. Reference-image generation and membership access are separate considerations.
| Duration | IR disabled | IR enabled | Both candidates |
|---|---|---|---|
| 4 seconds | 60 points | 65 points | 125 points |
| 8 seconds | 120 points | 125 points | 245 points |
| 15 seconds | 225 points | 230 points | 455 points |
At 2K, replace the base rate with 30 points per second and add the same five-point surcharge when enabled. A four-second 2K candidate therefore costs 120 or 125 points. The surcharge is not five points per second.
Use the points planning guide to put this comparison within the user's total budget. Active membership remains required for API use; points alone do not establish eligibility.
Record an intentional experiment
Keep the prompt, image reference, duration, ratio and resolution the same between candidates. Record the changed flag and task IDs in your application. This is an example evaluation record, not a payload accepted by the generation endpoint:
{
"shot": "cup-close-up",
"acceptance": ["hand stays on cup", "no extra cup appears"],
"fixed": { "seconds": 4, "resolution": "768P", "aspect_ratio": "16:9" },
"candidates": [
{ "context_ir": false, "operation": "cup-ir-off-01", "task_id": null },
{ "context_ir": true, "operation": "cup-ir-on-01", "task_id": null }
]
}
Each candidate needs its own generation request and idempotency key because its parameters differ. An interrupted retry of either candidate should preserve that candidate's original key and parameters.
Review outcomes without inventing a benchmark
Wait for both task states, then download successful outputs through their authenticated content endpoints. Play the entire clip, including its beginning and ending. Inspect the acceptance criteria at normal playback speed and pause at moments that matter.
A single pair cannot establish that enhancement always wins: generative results vary, and a changed result is not automatically an improved one. Keep the observed result tied to the exact shot. Do not turn an internal preference into an advertised quality guarantee.
If your concern is compatibility of the requested action rather than detail, use the positive-prompt compatibility guide to revise the instruction deliberately.
Handle incomplete evidence before setting a default
HTTP acceptance is not proof of a finished clip, as MDN's explanation of HTTP 202 makes clear. Inspect task status and any returned error instead of treating a response code as a quality result.
If one candidate fails, record that outcome rather than comparing the other against an absent video. If the upstream result is unknown, preserve the operation identity and seek reconciliation before another submission. Adopt enhancement for the tested use case only when its reviewed result justifies its extra cost; leave it disabled when the evidence does not support that decision.
Related tasks
- Imagild Points, Resolution and Jobs: Plan a Creation Before You Spend
- Why Does Adding “Do Not” Make My AI Video Less Predictable?
Sources and scope
- Imagild API integration guide: Actual request contract, supported references, asynchronous tasks and active-membership access.
- Imagild model hub: Current H3 duration, resolution and ratio choices, video point rates and optional enhancement surcharge.
- MDN HTTP 202 Accepted: HTTP acceptance is distinct from completed processing; no Imagild availability or quality claim.
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.