LoRA trained on the animefull-final model using the official artwork of the mainline Pokémon games. I'd personally recommend using this the most with AnythingV3.
For full body shots I'd suggesting using a resolution of 448x896, as this was the largest bucket in the dataset. Most resolutions still work quite well though and which you want to use can depend a lot on the pose/character you want to depict.
All artwork has been tagged with their respective generation (eg. generation 1, generation 2, etc.). Allowing you to get different styles, occasionally there's a bit of bleeding (mostly at gens 3 and 4) but its fairly consistent otherwise.
As for upscaling, I recommend using something like the anime ESRGAN models, rather than the default latent upscaling. As the Pokémon artstyle doesn't have a lot of details any new stuff added usually looks bad.
Description
FAQ
Comments (4)
Good model. Make an XL version please. I only use XL.
I've been considering it but haven't really kept up with all the latest happenings at all and am completely in the dark of how training for XL works, I used to do these models in google colab as well because I don't have a powerful enough GPU but I doubt that'd work these days. I might try to look into it but I wouldn't count on it honestly.
@dubby Ok. Thanks for the reply.
@dubby The on-site lora maker is actually pretty good and cheap since it goes off buzz, and honestly, going from 1.5 to Pony/XL isn't a huge leap.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.