･˚‧･+‧₊‧.°.⋆.🫧.•˚₊‧⋆:｡+.･
🩵 It's me again with another cute/pastel iridecent style LoRA xD. I wanted to make something that I'm really into since it's gonna be my last LoRA for the event 😆.
💜 A very special thanks from me to everyone for your support all this time 💝. I will try my best to make more and better LoRAs in the future 😊.
🩷 As usual, I hope you'll like this LoRA and can't wait to see your awesome images!
･˚‧･+‧₊‧.°.⋆.🫧.•˚₊‧⋆:｡+.･
Description
💌 Standard weight is 1 but feel free to adjust the numbers to your liking.
💌 Please avoid using too many LoRAs at once since it might mess up with the results.
💌 Results may vary on different checkpoints.
FAQ
Comments (10)
super pretty. will there be a pony version as well?
Thank you 🩷 Umm I've never trained a Pony lora before so Im not sure how different it'd be 😥 I'll look into it tho 🤔
it works so well. wow, awesome job!! ♥
Thank you 🩷🩷🩷
This lora is so cool, thank you !
Woa thank you too 🩷 I'm glad you like it 😊
Didn't do an extensive test, just a dozen different attempts yet my impression is that this World Morph doesn't really like Pony/IL models. I get a sludge with those if anything, XL/Animagine character LORAs seem to work fine, albeit converted to 2.5D, but that's the expected. I look forward to the glittering scale wear I can put on the characters!
EDIT: Changed my comment up to be more accurate.
Where's the advice?!
Interesting LoRA.
Had mixed success so far. Not able to constraint the effect to main subject. (Noob disclaimer).
I believe that there is a side effect that female characters frequently develop wings. Was it intended?
I am using simple sdXL_v10VAEFix + sdXL_v10RefinerVAEFix workflow locally.
A WARNING to Illustrious users.
This LoRA seems to modify CLIP text encoders in a way incompatible with Illustrious-derived models. The ordinary ranges of LoRA strength (e.g., 0.7-1.0) for text encoders don't work. In particular, the image starts distoring around strength 0.3 on the original Illustrious XL models. Any higher strengths are completely unusable. Similar behaviors are observed in all Illustrious-derived models in my test run, although the distorting point varies in ranges 0.2 - 0.5. (I tested 5 Illustrious finetunes by different creators.)
A workaround on ComfyUI is to keep the `strength_clip` value very low, e.g., 0.20, and adjust the strength of LoRA only with `strength_model` value. (You may need to modify your workflow to use the standard "Load Lora (Model and Clip)" node in Comfy Core to do so.) You could try slightly higher `strength_clip` value, depending on your base model.
If your generation tool doesn't allow specifying separate LoRA application strengths for CLIP and U-Net, you need to keep the unified LoRA strength very low, in exchange of limiting the effect of this LoRA. Another trivial workaround is not using this LoRA or Illustrious-derived models.
Enjoy! 😓