Initial version of Chimera XL, focusing on (un)realistic creatures, hybrids, monsters.
The training data had Gen1 pokemons (just their names, no type tagging), some LoL characters (Ashe, Akali, Ahri, Katarina, Star Guardian Jinx, Nidalee, etc), some OW characters (mercy, dva, black cat dva, tracer), species (tauren, navi, smurf, succubus, draenei, lamia, turian, asari, krogan, skrull), various wereanimals and hybrids, fantasy creatures (undead dragon, harpy, dryad, unicorn, centaur, humantaur, kobold, goblin, floran, avian, fairy, arachne, slime, griffin), regular animals, and various LUTs (see LUT Diffusion LoRA).
I used Dreamshaper Alpha2 as a base.
Description
further trainings, also added some synthetic data to reinforce some emerging concepts, both from previous versions and from dalle3
seems to be coherent and versatile enough for a new version
also gathered some transformation scene images, in hopes that it helps with better understanding morphing between concepts, sometimes that tag also works, but isn't reliable enough yet; prefers wide images
dataset was tagged like: X to Y, transformation scene; X turning into Y; X transforming into Y
experiment with Xtaur
FAQ
Comments (9)
Hi, there seems to be a problem with the VAE inside the 2.4 model. It gives these blue dots in shadow areas. Mostly noticeable in corner of mouth and dark areas of the hair. It might come from the base model you are using, but consider swapping it out for one without these numerical issues for the next version.
thank you, i'll look into it 👌
I have the same problem as @pwsoft . For me it was SassyGirl model (i like the model). When i mixed with others, the blue dots started to appear. Thank you for your Chimera models. v17 was really impresive. Keep at good work. :)
Newest Version doesn't work at all for me. Only gibes me noise.
Ol found the Problem. Doesn't work with the some Samplertypes in ComfyUi. Great Results so far!
Great version!
There is an issue I'm having with artifacts though. In all of my outputs I am seeing grey circular speckles throughout the image. It seems as if they are targeting specific parts of the output as well.
I am using ComfyUI. I am using your sampler settings as well that are shown on this model listing. I will attach my ComfyUI workflow onto my images down in the Gallery for you to look at and maybe even help fix.
Great work as always though.
thanks, i was able to reproduce the issue, and it seems like my newer iterations got rid of that. needs some more time testing it though
While wier is fixing that, you can still use the current 2.4 model if you use a different VAE. As the problem is only in the VAE. So in ComfyUI only use the MODEL and CLIP outputs of the loaded Chimera model. Then add another loader block with any other SDXL model, that doesn't have that problem and use the VAE output from that to connect to your VAE Decode.
There is also a VAE loader in ComfyUI in case you already have a VAE file. Hope that helps.
@mack @pwsoft i uploaded a new version, rebaked the base sdxl VAE, and diluted it a bit. please let me know if you still run into these issues
@wier Yeah that indeed seems to have solved it for me. Didn't spot any blue spots using 2.5. Will have to experiment some more with the model now that it functions correctly, so I can properly write a review. Initial results I get are promising though.