Pokémon Trainer Sprite Generator
LIKE THIS MODEL? GIVE A ❤️
Trained on Pokémon sprites of 96x96 and BLIP image captions.
Quick Tips
No Trigger Word Needed: This model does not require a trigger word.
Art Styles: You can use the tags
gen1andgen2to generate different art styles.Trainer Recognition: The model can recognize some trainer names, such as Ash and Bruno.
This LoRA has been tested with Dreamshaper and RealisticVision, but I belive that it should work well with other models too.
Model Versions
This Pokémon Trainer Sprite LoRA is available in two different versions: 768x768 and 512x512. Both models are trained on images upscaled from their original 96x96 size and are specifically designed to produce sprites at their respective resolutions. The choice between these two versions allows users to customize their experience based on their specific needs.
For the best results, we recommend using the 768x768 version. Set your desired resolution to 768x768, and then downscale the result 8 times to achieve a 96x96 sprite. This downscaling process can be performed using any image editing program, but for a smoother post-processing experience, we recommend using a plug-in available at (https://github.com/mrreplicart/sd-webui-pixelart).
Enjoy your Pokémon Trainer sprite!
Description
This version train on 768x768 resolution which give more accurate pixelate result than 512x512 version.
We suggest setting your image resolution to 768x768 and then downscaling it 8 times to 96x96 for optimal results. Any image editing programs will work, but we recommend using plug-in for post-processing. (https://github.com/mrreplicart/sd-webui-pixelart)
FAQ
Comments (8)
Are the results you shared tweaked? Because I don't seem to be able to replicate them, maybe I'm missing something
Read the "About this version" to get more detailed info on the model. It does work, and it helps if you use a pixel extension along with the model.
@LindezaBlue I've had success using the method Surdjak suggested in his review.
I didn't know he Pixelization extension, seems good, I'll try it. Thanks!
@JohnAiArt glad I could help :D
I have the same problem. Can someone explain me why (I'm new to SD)
* downloaded the same checkpoint and Lora (compared hash sums)
* copy pasted generation data and used auto-fill feature of the interface (including same seed, same size, same checkpoint, same positive prompt, negative prompt empty as in orig)
Nice work. Is it possible to create a Pokemon character of an existing person from a photo (img2img)? Is there a tutorial?
Can the current training method achieve the model of trainers' overworld sprites?
This LoRa cannot generate overworld sprites since it didn't train for. I have tried training an overworld model. The result didn't come out well yet.