Ayfri/GPT-Images

Ayfri/GPT-Images is a solid “bring your own OpenAI key” image and video generator. It does a good job staying focused: generate images, attach references, tweak model/quality/size settings, keep a local history, and show rough usage costs without trying to become a full design suite.

The best part is the architecture choice. Keeping the API key in the browser makes sense for this kind of personal tool, and the project is honest about that tradeoff. I also like the split between IndexedDB for metadata and OPFS for larger media files; that shows the maintainer has moved beyond demo code and thought about what happens once people actually build up a gallery.

The UI is practical too. The live app exposes useful controls without feeling overloaded, and the pricing table is a nice touch because image generation costs can be confusing. The repo also has good momentum, with 100+ commits and support for newer image/video model options.

The main things I’d tighten are documentation and project hygiene. The README still mentions pnpm, while package.json now points to Bun. The license is also inconsistent: GitHub shows GPL-3.0, but the README says MIT. I’d fix those first because they affect anyone trying to use or contribute to the project. After that, adding a small CI workflow for bun run check and bun run build, plus tests around pricing and storage migrations, would make the project feel much safer to build on.

Overall, this is a useful, thoughtfully built client app. It already works as a real personal tool, and with a bit more polish around docs, licensing, and automated checks, it could also be a strong reference project for SvelteKit apps that talk directly to AI APIs.