I integrated a native image generation model directly into a Java application using Quarkus and the Java FFM API.
No Python stack, no sidecars, no external services.
Just in-process inference, explicit memory ownership, and predictable behavior — even on Apple Silicon.
Write-up with full code and caveats:
https://www.the-main-thread.com/p/java-local-image-generation-quarkus-ffm-flux
