
Compare every important word in an API-generated image against the approved copy before delivering it. A successful task or a higher resolution tier does not guarantee spelling, punctuation or correct placement. Keep the expected text as a separate application record so the reviewer can check the actual saved image rather than rely on memory.
Key Takeaways
- Put the exact desired wording in the prompt and retain an approved text checklist.
- Treat OCR as optional external assistance, not proof that all lettering is correct.
- Separate a new generation from a downstream typesetting correction.
Freeze the copy that must appear
For a portrait caption, a name, role and location can each be an independent requirement. A reviewer should know whether capitalization, line breaks or punctuation may change.
This application-owned checklist is an example, not an Imagild request body:
{
"asset": "designer-profile-card",
"expectedText": [
"Maya Chen",
"Product Designer",
"London"
],
"caseSensitive": true,
"reviewDecision": "pending"
}
Use the Imagild API guide for supported request fields. There is no separate expectedText or OCR parameter in image-generation JSON; carry the creative wording in prompt and keep this checklist in your application.
Give the model a reviewable instruction
Describe the text panel and provide its exact wording. Identify what should remain from the reference, such as the portrait, outfit and background. An image-reference request needs at least twenty trimmed characters of meaningful instruction.
The reference scope guide helps separate a local caption edit from replacing the whole scene. Avoid silently substituting a demonstration name or inventing a job title the user did not supply.
Check the model hub for current resolution choices and points. Paying for 4K changes the request tier and cost; it is not a lettering guarantee.
Inspect the final crop at its delivery size
Download the successful result with authentication, then review the source file and its intended presentation crop. Read each line slowly, including small secondary text. A large preview can hide that the final mobile card makes the role unreadable.
Compare the result with the approved checklist rather than another generated image. Look for missing letters, repeated words, altered names, added text and visually plausible but incorrect characters.
The resolution and job guide supports budgeting for this review process. Requested pixels and a successful job state are different from copy approval.
Use OCR without giving it final authority
If your own pipeline includes an OCR tool, use its output to flag possible differences for review. Imagild's public API does not provide that OCR stage.
Tesseract's quality guidance explains that image conditions can affect recognition. A matching OCR string therefore does not prove that tiny, stylized or distorted lettering is visually acceptable. An OCR mismatch may also be a recognition problem rather than a generation error.
Keep the recognized text, reviewed image version and human decision together. Do not silently normalize away case or punctuation that the approved copy requires.
Choose a correction that addresses the defect
If the composition is sound but exact copy is essential, a separate layout or typesetting step in your own production workflow can be the appropriate correction. That is not an additional Imagild endpoint or an undocumented mask option.
If you choose a new generation, approve its revised prompt and extra points, preserve the earlier task and create a new operation identity. Do not confuse that retake with an identical network retry.
Pass the asset when the actual delivered crop matches the approved wording and remains legible at the intended size. Record unresolved text defects honestly rather than labeling the image ready because the API task succeeded.
Related tasks
- Should an AI Reference Replace an Object or Bring in the Whole Scene?
- Imagild Points, Resolution and Jobs: Plan a Creation Before You Spend
Sources and scope
- Imagild API console: Actual request and task envelopes, strict parameters, references, membership, points usage and authenticated result retrieval.
- Imagild model hub: Current advertised model capabilities, resolutions and generation point rates.
- Tesseract output-quality guidance: OCR accuracy depends on image conditions, so external recognition does not replace visual copy approval.
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.