Purpose
Why the library exists
Product mockups, sample accounts, chat interfaces, and prototypes often need believable profile pictures without implying that a real customer, employee, patient, student, or reviewer exists. This library provides synthetic alternatives and explains how to create a custom avatar in ChatGPT when a ready-made image is not suitable.
Generation method
How new avatars are produced
-
Generate fictional subjects
New portraits added to this project are created with Codex's built-in image generation tool. Real-person photos and scraped profile images are not used as gallery source material.
-
Keep batch categories consistent
When a multi-image sheet is used, every portrait on that sheet has the same intended visible heritage category and gender presentation before individual images are cropped.
-
Crop and describe each image
Selected portraits are prepared as square 512 by 512 pixel files. Metadata records the visible presentation, age range, style, expression, and a plain-language caption.
-
Review the complete batch
A contact-sheet overview is used to check for repeated face templates, overly uniform hair or styling, weak crops, and visual details that conflict with assigned tags.
Metadata policy
What the visible-presentation tags mean
Heritage and gender-presentation labels describe the intended fictional subject and visible presentation of an image. They are search filters, not claims about a real person's identity. If an output is ambiguous or visibly conflicts with its metadata, the record should be corrected or the image retired instead of being left under an inaccurate category.
Maintenance
How guides and corrections are maintained
AI Avatar Library maintains the pages, prompt examples, image metadata, and review process. ChatGPT feature claims are checked against the official OpenAI Images documentation linked from the relevant guide. Substantive corrections update the page review date; dates are not changed merely to make old content appear fresh.