Support for Flux PuLID is now baked into AIProxySwift. You can use this model to generate headshots without a separate training step. The image-to-image generator works on a single image that is sent up as part of the inference request.

In our testing, the PuLID generator is fast (~ 10 seconds) if not completely accurate. If some variance in output is allowed in your application, this could be a great model for you.

To use Flux PuLID with AIProxy, use AIProxySwift version 0.23.0. or later. You can copy the sample snippet straight from the PR that added support. Make sure you configure your AIProxy service to proxy requests to the Replicate API: