
Imagild exposes three canonical image model IDs: gpt-image-2.5, nano-banana-pro, and nano-banana-2.1. Use the authenticated GET /models response to populate a selector, then submit the exact returned ID. Similar names on another platform are not evidence of an interchangeable request format or identical output.
Key Takeaways
- Filter the live catalog by
kind: "image"; do not turn provider aliases into additional choices.- Each current image choice offers 1K for 10, 2K for 15, and 4K for 20 points.
- Store the selected model with each operation so later UI changes cannot relabel an existing task.
Read the catalog before building the dropdown
Start in the Imagild API console with an active membership and a valid API key. Purchased points fund generation, but a points pack alone does not grant API eligibility. Keep the key on your backend.
This read-only discovery request uses the same Base URL as generation:
curl --fail-with-body \
'https://imagild.com/api/relay/v1/models' \
-H "Authorization: Bearer $IMAGILD_API_KEY"
The JSON envelope contains data.models. Entries include a model id and kind; current image entries also describe supported resolution tiers and points. The endpoint returns configured, available choices. Do not assume a model remains selectable forever because an old screenshot showed it.
An application can turn that response into a small selector without inventing aliases:
const response = await fetch(
"https://imagild.com/api/relay/v1/models",
{ headers: { Authorization: "Bearer " + process.env.IMAGILD_API_KEY } }
);
if (!response.ok) throw new Error("Model discovery failed");
const { data } = await response.json();
const choices = data.models
.filter(model => model.kind === "image")
.map(model => ({ id: model.id, resolutions: model.resolutions,
points: model.points }));
console.log(choices);
This is server-side JavaScript. A failed catalog request should produce an actionable discovery error, rather than silently inserting a guessed model.
A routing ID is not a compatibility certificate
Model discovery answers which identifier the Imagild endpoint accepts. It does not establish that an official vendor SDK can be pointed at this Base URL unchanged. For example, OpenAI's image reference describes its own image request and response fields. Imagild uses asynchronous tasks and its own strict image schema.
Keep your adapter explicit: model, prompt, resolution, and, when needed, one owned reference selector. Fields such as size, n, seed, and response_format are not accepted image options here. Neither is image aspect_ratio.
This distinction is a useful design discipline across platforms. Runware's model schema documentation likewise makes input support discoverable at the model contract level. Its field names are not Imagild parameters; the transferable lesson is to validate against the receiving service.
Make model changes visible to the user
Show the selected ID and resolution near the submit action. A compact quote such as “Nano Banana Pro · 2K · 15 points” communicates the paid decision more clearly than a list of nearly identical internal channel names.
For a comparison screen, keep the prompt and resolution fixed, then create a separate operation for each selected model. Label outputs from their task records. Do not claim a quality winner from the catalog, since availability and price do not measure text fidelity, identity preservation, or latency.
Keep existing jobs attached to their original model and task ID. Changing the dropdown should affect the next submission, not rename a pending result or trigger an unrequested generation.
Refresh the choice without rewriting history
Refresh discovery when the user opens generation settings or when a stale choice is rejected. Re-check the current model market for the visible offer. Preserve the original operation record and point quote for work already submitted.
For the wider planning workflow, see resolution and job budgeting. If a selected request fails, follow the failed-job next-action checks before making a new paid operation. A clean selector offers a small set of real choices while preserving an accurate history of what the user actually requested.
Related tasks
- Imagild Points, Resolution and Jobs: Plan a Creation Before You Spend
- What Should You Check Before Retrying a Failed Image or Video Job?
Sources and scope
- Imagild API console: Imagild Base URL, authentication, async task contract and active membership prerequisite.
- Imagild model market: Current canonical image choices and resolution pricing.
- OpenAI image generation reference: OpenAI maintains its own image request and response contract; not proof of Imagild SDK compatibility.
- Runware Model Schemas: Model-specific schema discovery and validation as a distinct integration discipline.
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.