Just wanted to add that ollama doesn't support text-to-image yet, ragdoll implements stable diffusion's txt2img separately. I tried that Phi one, it was very bad, not even close to ready for primetime. My favorites from ollama so far are mistral for most uses and alfred for highly instructional use cases. I can't even try some of them because of the requirements, but some of the apparently high-end ones I have tried just aren't good - unusably so. Mistral is pretty awesome, glad they are fully open.