This is an old SD1.5 lora I made several years ago. I wanted to upload it in the meantime while I work on porting it to SDXL. The training is a bit lopsided, so some characters are weakly trained, mainly Eleanor and also Theodore and Jeanette are a little weak.
You'll want to mostly use 0.7 weight. You can bump it up to 0.8 if the accuracy isn't good enough, and in cases like Eleanor, you may need to use 1.0 weight along with supporting tags to get the shape down, and then inpaint or img2img afterward with a lower weight.
Tags are as follows:
CGI is a mandatory tag.
Use the first name of each desired character. Such as Alvin, Simon, Theodore, Brittany, Jeanette, and Eleanor.
Add Chipmunk or Chipette according to the desired character's gender. It might still work otherwise, but that's how the training data was tagged.
I recommend using hires fix on a 1.5 multiplier. Or just generate at 1536 for one or both of the width and height.
Description
Initial Version - I recommend 0.7 weight for the most part. Some characters did not get trained as thoroughly, so you may want to use 1 strength for them and supporting tags to get a shape, then hires fix, img2img, or inpaint on 0.7 weight to refine it.
Comments (2)
This is amazing and I've been waiting so long for something like this, however it's really hard to not get monstrosities with this one. Will you make a more updated version later on?
Realistically your best results will come from inpainting images or using hires fix to refine them. It is SD 1.5 after all, which is not very good for stuff beyond solo art, and even solo art is spotty. Apart from that, just look at my preview images to get an idea for how to tag it.
As for updates, this will never be updated. It is obsolete. I am slowly working on porting it to Pony Diffusion.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.