Codex Image Studio
Generate and edit production-ready raster assets for projects.
The documentation clearly separates built-in tool use from explicitly requested CLI fallback, forbids automatic fallback, requires OPENAI_API_KEY via the environment, prohibits pasting secrets, and specifies non-destructive project output handling. Deductions apply because images are sent to an external image service while data flow, privacy retention, cost confirmation, and moderation boundaries are not fully disclosed; explicit per-request user confirmation and verifiable source attribution are also absent.
The workflow, mode boundaries, image-input roles, output conventions, parameter validation, file checks, and some retry behavior are documented. The score is limited because this is a static review with no key-path tests or CI evidence; built-in tool behavior, dependency availability, and abnormal-input coverage cannot be reproduced from the supplied files.
The audience, use cases, generate-versus-edit rules, non-fit cases, and distinction between reference images and edit targets are reasonably clear, supporting semantic invocation. Deductions apply because Chinese-language prompting and user experience are not addressed, mainland-China network reachability is not established, and the CLI depends on an external API.
The material has strong information architecture with progressive disclosure, mode rules, a decision tree, prompting guidance, examples, limitations, dependency notes, and output conventions. Deductions apply because versioning, changelog, named maintenance responsibility, and update path are not clear; license metadata is unknown, and the repository README states that the repository is deprecated.
The skill clearly covers image generation, editing, batching, and multiple asset categories, with direct image-output and project-save guidance. The score is limited because no representative outputs or third-party execution evidence are supplied, so image quality, text accuracy, edit fidelity, and marginal benefit over manual alternatives cannot be verified statically.
The supplied skill files, references, and script expose auditable rules, parameters, and failure paths, with generally consistent documentation. Deductions apply because there are no committed tests, CI results, execution logs, or independent review evidence; conclusions rely primarily on author-provided static material.
- Generation and editing involve an external service, potentially sensitive images, and API costs; confirm data handling, billing, and content policies before use.
- The repository README marks the repository as deprecated, so current maintenance ownership, version status, and update path require separate confirmation.
- No evidence establishes Chinese-language support or mainland-China network reachability; affected users may need additional environment configuration.
- No tools or scripts were executed in this assessment, so output quality, dependency installation, and failure behavior remain unverified.
What it does & when to use it
Imagegen is the image-focused skill in the openai/skills repository, covering raster tasks such as photos, illustrations, textures, sprites, product mockups, UI mockups, and transparent cutouts. Its preferred path uses the built-in image_gen tool and does not require OPENAI_API_KEY; the scripts/image_gen.py CLI is an explicit fallback only. The instructions cover generation, editing, batch variants, structured prompting, image-role labeling, non-destructive edits, and saving project-bound assets. The repository README states that the repository is deprecated and directs users to the OpenAI Plugins repository for current examples.
Generates new raster images from prompts or edits existing images visible in the conversation; local edit targets should first be loaded with view_image. It chooses generate or edit intent, single-asset or multi-variant execution, and a defined image-use taxonomy with structured prompt fields. When CLI mode is explicitly requested, it uses scripts/image_gen.py with generate, edit, or generate-batch, requires a locally configured OPENAI_API_KEY, and instructs users to copy project-bound outputs into the workspace.
- A website or product team needs a landing-page hero, product photograph, or packaging mockup and wants a raster asset generated from a structured brief.
- A game developer needs concept art, sprites, or several asset variants for one task.
- A designer needs an app or website UI mockup, wireframe, or infographic with specified layout and text.
- A user needs a background replacement, object removal, lighting or weather change, or transparent-background cutout.
- A user needs to preserve a person’s identity, pose, or a product while changing only a specified visual element.
Pros & cons
- Covers common raster generation, editing, and variant workflows.
- The preferred built-in path does not require the user to provide OPENAI_API_KEY.
- Provides concrete guidance for image roles, editing invariants, and structured prompts.
- Clearly separates raster-image work from tasks better handled with SVG, HTML/CSS, canvas, or native code.
- It relies on Codex-oriented tools such as image_gen and view_image, so other clients may require adaptation.
- The CLI fallback requires OPENAI_API_KEY, Python, uv, and the openai Python package.
- No test suite, platform support matrix, or empirical image-quality evaluation is provided.
- The repository README marks the repository as deprecated and points maintenance guidance toward the OpenAI Plugins repository.
How to install
imagegen is located at skills/.system/imagegen/. The README says that .system skills are automatically installed with the latest version of Codex, so separate installation is normally unnecessary. The repository is deprecated; current skill and plugin examples should be obtained from the OpenAI Plugins repository. The individual skill license is expected in LICENSE.txt inside the skill directory, but its contents are not provided here.
How to use
Ask for an image generation or editing task and specify the intended use, subject, style, composition, lighting, text, and constraints; the default execution path uses the built-in image_gen tool. For a local edit target, load the image first so it is visible in the conversation. To use the CLI, explicitly request CLI mode, set OPENAI_API_KEY locally, and use scripts/image_gen.py with generate, edit, or generate-batch.
Compared to similar skills
The source explicitly contrasts raster generation and editing with SVG, vector icon systems, HTML/CSS, canvas, and other code-native output. Choose those alternatives when the desired result is deterministic vector or code-native content.