AIProxy now supports Fal.ai's fast flux trainer end-to-end. This means you can train a headshot model using images available to your iOS or macOS app, and then generate images from the trained result. In our experience, training takes around 2 minutes and cost $2, and inference takes around 10 seconds.

Here is an image I created of myself floating over Monument Valley, Utah. In this proof of concept I only used a single training image (my twitter avatar) and I was seriously surprised to see the quality of the result.

Snippets for uploading training data to Fal, training the model, and performing inference are all available in the PR that added support