Hi!
We just released our open-source package MediaGenerationKit: Introducing MediaGenerationKit: Hybrid Media Generation for Production Apps
It is based on our Draw Things inference stack, so local media generations will be fastest (see comparison: Metal Quantized Attention: pulling M5 Max ahead with Int8 matrix multiplication ). It also allow you to offload the generation task to your self-hosted Cloud or to Draw Things Cloud Compute backend for less powerful devices.
Let me know if you have any questions!