v1.1 update: No longer a LoCon model and so no longer requires any extra extensions to work!
Coco Bandicoot from Crash Bandicoot!
Use the tag "coco". Optionally you can use "furry, overalls, white shirt, goggles on head, flower in hair", for various parts of her outfit.
Works well with most models I have tested it with. Have fun!
v1 is a LoCon lora and requires this extension to work with the latest webui.
Description
Only "coco" is required in the prompt, but adding "furry" as well can improve your results.
FAQ
Comments (13)
I love the toy preview pic!
Mage says keep making great stuffz, thanks!
I've got an error. All other LoRAs that I use (Amalia, Gwen, ArcaneStyle, etc.) work great. I tried your mentioned 'fix' - reverting to specified commit, your LoRA works, but everything else working weird with it.
I was having some issues with that version and ended up rolling back a little further to version: 2f0181405f25e1448a55697081e380020fe8c68d
I need to update the description.
I think the problem is that I used the LoCon LoRA method in Kohya, and the latest version of auto1111 has issues with that. So alternatively, I recently saw that Civitai recommends installing this extension for LoRA's using LoCon. I haven't tested this myself but in theory it should make it work with the latest version. https://github.com/KohakuBlueleaf/a1111-sd-webui-locon
Ok, I have now tested it. Installing the extension suggested by Civitai (https://github.com/KohakuBlueleaf/a1111-sd-webui-locon) does in fact fix issue and lets you use this lora as normal on the latest version of the webui.
@Puffin Thanks, it helped ^_^
@Puffin For me, the latest version of SD WebUI also caused problems even with the newest LoCon Extension (I tested Elora, Marie Itami and Coco). However, I am using an AMD GPU and thus a modified version of A1111, which might have caused this. I went back to an older version and now everything is fine again. Great models btw!
Tawna old style not crash 4 be cool
this looks incredible! any training tips?
I highly recommend following this guide, it's dense but the information is the best I've found: https://rentry.org/59xed3#preamble
The only thing I do differently is I crop all my training images to be 1x1 and resize them to 1024x1024 before training. This makes it so they aren't spread across a bunch of buckets during training and I find improves the LoRA.
i wish this was a normal lora
I intend to retrain my LoRA's using the default method since LoCon is no longer natively supported. I'll do this one first, so keep your eyes open for an updated version in a day or so!
@Puffin i really hope so because this is one of the best looking loras that i really want to use
Just uploaded v1.1, very similar to v1 but is now a normal LoRA!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.