Here's a model for an early 90's Mazda Miata, as requested by someone in the comments. I'd suggest keeping the weight around .8 for the best results. I'm planning on releasing a v1.1 tomorrow with some additional features and new training. Right now, you can prompt it for "convertible top up" "convertible top down" and "Hard top", and it will react fairly well. Sometimes it has a hard time understanding the convertible top and the role bars of some of the cars. You can also try to prompt if you want to see the front side or back side. The version tomorrow you should be able to prompt it to flip the headlights up and down! All images were trained at 768x768 as well, so you can go a little bit higher resolution. For the checkpoint, I've actually started using this one for all of the images and it's been giving me good photo real results: Dreamlike Diffusion: <a target="_blank" rel="ugc" href="https://civarchive.com/models/1274/dreamlike-diffusion-10">https://civarchive.com/models/1274/dreamlike-diffusion-10</a>

Files with SHA256: eaed1c0a125954926174e4b3e1a57d5b1cfcfc39393dfb5158bc7f354e1b3648