Skip to main content
Technology & EngineeringImage Generation Services29 lines

OpenAI GPT Image Generation and Edits

Choose a GPT Image generation or editing path, preserve approved visual references, and verify the returned image files and requested changes.

Quick Summary8 lines
Identify the intended output, dimensions or aspect ratio, format, transparency and visual references. Confirm the model and endpoint capabilities in the OpenAI image generation guide. Use the Images API for straightforward generation or editing, and consider the Responses image generation tool for conversational edits when that matches the existing application. Record the selected API, exact model and input files without putting secrets in a prompt or output log.

## Key Points

- [OpenAI image generation guide](https://developers.openai.com/api/docs/guides/image-generation)
- [OpenAI image generation tool](https://developers.openai.com/api/docs/guides/tools-image-generation)
skilldb get image-generation-services-skills/openai-gpt-imageFull skill: 29 lines
Paste into your CLAUDE.md or agent config

OpenAI GPT Image Generation and Edits

Identify the intended output, dimensions or aspect ratio, format, transparency and visual references. Confirm the model and endpoint capabilities in the OpenAI image generation guide. Use the Images API for straightforward generation or editing, and consider the Responses image generation tool for conversational edits when that matches the existing application. Record the selected API, exact model and input files without putting secrets in a prompt or output log.

Prepare the image brief

For generation, describe the subject, composition, lighting, materials and delivery constraints. For editing, identify the primary reference and state the requested change and details that should stay consistent. Order multiple inputs deliberately and verify the selected model supports the desired number, formats and masking behavior. Use approved references and retain untouched originals. If identity or established IP fidelity is part of the project, compare a representative test against authoritative references before preparing a batch.

Describe requested text exactly and check whether the receiving use requires exact glyphs and spacing. A separate text layer can provide reliable lettering in a production design. Specify transparency explicitly only where supported by the model and output format. Record the requested background, size, quality and file type.

Handle outputs as artifacts

Read the selected endpoint's response shape and save the image bytes or decode the returned base64 into a new file. Validate image format and dimensions by opening the saved artifact. Inspect alpha if transparency is required. Keep a local job receipt with a sanitized request summary, reference IDs, requested settings and returned metadata. Distinguish settings requested from model or quality actually reported. Do not log credentials, private input media or full response payloads unnecessarily.

Review and iterate

Compare the returned image with the brief and approved references. Inspect silhouette, geometry, palette and text at intended size. For an edit, compare areas surrounding the change and check unrelated details for visual drift. If the change misses the intended region, refine the prompt or supported mask and retain the approved image as the next reference. Log each accepted or rejected revision. Preserve the original generated file before optional resizing and avoid adding enhancement that the user did not request.

Sources

Install this skill directly: skilldb add image-generation-services-skills

Get CLI access →