Update: Uploaded a version that ungroups the AYS Custom Sampler to make it easier for troubleshooting.

You'll need Ollama installed and a LLM that's good for analysing images and one that's good for embellishing prompts.

Load image is fed into first LLM and asked what the image is and to make a prompt from it. That is then fed to the second LLM and asked to improve the prompt. There's also normal positive and negative prompt boxes for manual guidance.  

The general image gen workflow started off after following this video https://youtu.be/nZx5g3TGsNc?si=A6J565spdpAGGhlD, but then tinkered and it's quite different now. There's a standard KSampler and a Face Detector for one output and a grouped set of nodes for AYS, PAG, Auto CFG, Custom Sampler and a couple of others which then goes to it's own Face Detailer.

It's all in different groups and there's a group bypasser in the middle, so you can easily switch off anything you don't want.

Update, added a model uspcaler and SUPIR. It's set to a 2.5x upscale as that's as high as my 2080Ti will handle, if you have a card with more VRAM you should be able to increase the upscale further.