I'm convinced it's just a short time more until we see sprite sheets being generated entirely with AI, and that's the perfect excuse to "do it tomorrow", but I'm going to proceed today with the tools I have.
I've been talking about getting a new GPU for years, since 2019, and with NVidia announcing that there will be no new lineup of GPUs this year, along with the chip shortage and skyrocketing RAM prices, I made an impulse decision and bought a GPU. It arrives today.
I spent most of my time last week reading up on the nature of models and Lora's available for them, how they're trained, and how/why certain keywords are necessary to get them to run correctly. It was fun! And I successfully generated several images using my own prompts. I also changed up many of the parameters just to experiment because it's fun.
But I learned that my dear old GTX 1070 Ti just won't work efficiently. It only has 8GB of vRAM so I'm not even able to fully load big models into memory and processing time is abysmal. Hence the above mentioned impulse purchase that I have no regrets about.
But here are some of the images I created with the UmeAirt Lora trained on the Flux-1.dev model:
I saw a male cardinal in my backyard :)
Top comments (0)