Clover Ewing from Totally Spies
what I try to aim with her is to make her as flexible as it can with any model, meaning that I want it to not biased towards her TV Series/Cartoon style.
New Update for SDXL with Pony Diffusion V6 as base model
The picture preview is generated using AutismMix Pony, for known problem/more info, please refer below...
All the trigger word for SDXL Pony is the same as SD 1.5
Strength : 1 - 0.8~
Trigger Word :
clover \(totally spies\), blonde hair, short hair, blue eyesDefault outfit :
red bodysuit, beltThings you might want to add :
skin tight, red gloves, high heelsKnown Issues
SDXL Pony
v0.6
Play around with the strength/weight, I think it is "too detailed" with her mecha elbows going crazy all around, if at strength 1 that mechanical elbow goes too much, lower the LoRA strength
if you need even more flexibility for her to change the outfit, you can lower the LoRA strength (at 0.8 at most in my test) and also put red bodysuit in the negative prompt; please check the example images above.
I think I might need to rebuild the datasets, it's too unpredictable... just as some people said... trash in, trash out. BUT, it is still better than v0.5 in my opinion so... enjoy!
v0.5
Highly biased to her heart shaped belt, I'm trying to suppress it with lower strength, even add "heart" to the negative prompt, but still, the heart symbol showing, I guess because I'm still used the old tag for the 1.5 LoRA and it's causing the problem, I should tag the heart symbol too next time
Pony Diffusion as the base model pretty sure already have some totally spies character in it, seems like it's biased towards Sam, I suggest to add "green bodysuit" and "yellow bodysuit" on the negative prompt if you want to prompt her with the red bodysuit
SD 1.5
The elbow part of her red bodysuit (like that mecha thing) might random, the dataset not covering that part well
If you think the red bodysuit color not right, you can try to add the emphasize to it, ex : (red bodysuit:1.2)
If it still showing or biased on her red bodysuit while prompting her to have another outfit, try to lower the lora strength or add red bodysuit in the negative prompt
Her heart belt will also showing even while facing back/from behind angle, due to lack of dataset from her facing back, or the tagging that I did just too simple, maybe should tag it better like "heart-shaped belt" or something? I dunno, that'll goes for version 2
Support & Commissions
If you enjoy my work, please consider supporting me on Ko-fi or commissioning a custom LoRA. Every bit helps me create more!
For commissions and more, visit My Website
To anyone considering reposting or reprinting this LoRA to other sites : Please respect my efforts and do not remove any links, including my ko-fi link, from this page. Thanks.
Description
Update version 0.6, faarr different training method than version 0.5, overall, I think it is much better now. please read the model's detail for more information.
FAQ
Comments (11)
Fantastic Lora, but it keeps wanting to put green clovers on or around her since "Clover" is the trigger word, even when I put "clover \(totally spies\)".
Any advice on how to avoid that?
Oh, is it using the version 0.6-pony? I'm not sure but I think the problem shouldn't exist on that version...
But if it's still persist, maybe try to increase the lora strength or add more emphasize to her outfit
and if it keep doing so, I'll consider for other trigger word on next version...
@Barjoe01 That fixed it! Thank you so much! <3
Hey, it isn't usable on site any longer!
A bit of a joke-y ask: can a character pack of Brittany, Mandy, Alex, Sam, and Clover be made, such that the WHOOP catsuit has a LoRA where other colors can be picked (e.g. black and orange)?
It is kinda "Spy Chic" XD
It probably could but may take some high hardware power and time to be proper... but I'll see what I can do...
@Hoshikuchi will wait, but also aren't there ways to hack-around this (e.g. cropping, cross-training)? What are the limitation of characterpacks?
@TomLucidor I think it's more of between testing and re-training...
for character pack, I usually need for the dataset to be balanced, meaning that if I usually train 1 character for 1 - 2 hours, then 3 characters in one pack = 3 - 6 hours... and it's still not including the re-training, just in case I fcked things up... so yeah...
also because of my hardware too lol, it's only RTX 3060 12GB, could be faster by using paid google collab or such, but I don't have access to that for a moment
@Hoshikuchi Will keep notes on that
Hey, why was this archived?
LoRA model currently being archived for supporting fellow creators on the protest against civitai, you can use another site that might've re-posting this LoRA on their site.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.