You can try any model you want, however I really suggest using AnimagineXL 3.1 checkpoint for XL and AutismMix checkpoint for Pony.
Base model settings:
Steps: <30
Width: 896px
Height: 1132px
Sampler: Euler a
VAE: None
CFG Scale: 3 - 7
Clip skip: 2
Upscaler settings:
Upscaler: Latent (Nearest-Exact)
Upscale by: 1.5x
Denoising strength: 0.55
LoRA settings:
Weight: 0.7 - 1.0
Description
FAQ
Comments (13)
Aaah~ Tojo~~ It's weird that I can't use XL loras you made with Pony tho . I made XL lora once, and it works fine. I wonder why x_x
I'm sorry, I still have no idea why they don't work with Pony. Think it has to do maybe with the settings, most of the LoRAs I train I use PixAI's training tool, it isn't nearly as sophisticated as Civitai's or TensorArt's (but I have quite a bit of credits that I might as well use XD).
My GPU takes forever to train, but that's always an option. I might just start opting for it since I've become fascinated with Pony. For now bear with me just a little more haha.
I started this as a hobby but I'm glad many people seem to like my work. Be it for the characters or whatever but I see the support and I appreciate it.
@I_Tempest No need to feel sorry! I'm really happy that there's another person who focused on making male characters lora already! We need more ><
I myself only ever created Lora from google collab (Which cant be used anymore) and this website, so I dont really get what you mean, but I'm really happy that people appreciate your works! Here's hoping people know more about your account~ ><
Huh, it's weird. I can't even use pony lora that you made. Another user seems has the same problem. I wonder what happened? @_@
@RaidenFan86 I took a look into it and managed to make it work with other Pony based checkpoints, I tried Prefect Pony, Autism confetti, Autism DPO and pony diffusion v6 and didn't have any problem.
Which checkpoint are you using?
@I_Tempest I use animeboysv3. The loras dont appear on lora section of UI @_@ I tried editing the json file to convert it into Pony Lora (At first, It registered as 1.5 lora instead), but the result dont come close as the example. Rinne and Hiiro ones seems working fine tho @_@
Too much text warning; tldr at the end.
@RaidenFan86 I tried it with multiple LoRAs, not only mine but also from other creators and it seems a bit inconsistent. It works with some but doesn't with others, something else I noticed is it doesn't use pony's quality tags (score_9, score_8_up, source_anime, etc. This matters because none of the pony checkpoints so far have managed to make a workaround this hassle), so my assumption is that it's probably not trained in pony or some kind of mix between pony and SDXL.
Again, I'm no expert but I can't conclude anything else. Sorry I couldn't be of more help :c
I would however encourage you to try any of the other pony models, although they aren't mainly male-focused they provide very good results with the appropriate negatives and/or prompts. My go to is AutismMix Confetti but PrefectPony seems very solid (the style is a bit different though, but good)
TLDR; I assume that checkpoint is a mix of pony and SDXL or mostly SDXL, which causes inconsistencies. I recommend trying out other checkpoints like AutismMix Confetti and PrefectPony.
Thank you and hope you have a great day!
@I_Tempest Hi, no worries! I should say thank you for spending your time just to answering my problem. I really appreciate it!
I tried using Autism in my UI, but the problem still persist.
The lora still doesnt appear in the Lora section, and when i tried to "Force" it (By changing the 'json and change the settings into View all Loras), the result would just generate brown haired generic anime male. Not close to the example right here.
But strangely, when I uses Civit's online generator, it can works just fine!
Another user that created "Dash Baxter" lora seems facing the same problem.
I wonder what's wrong, do you remember changing something between creating Hiiro & Rinne's pony loras and this? Both of those works just fine, and Loras that I made using CIvit's trainer still works just fine in my UI.
The only other explanation is maybe there's something wrong with Civit's download and made the data corrupt. But we couldn't prove that, sadly :(
@RaidenFan86
I don't know if this will be confirmed by replying to past posts.
By default, this lora will not show up if you have selected the xl checkpoint in webui. However, if you select 1.5 checkpoints, it will show up in the lora. If you rewrite the json file for xl checkpoints, this lora will show up with xl checkpoints selected (some of this contributor's lora models for xl are in a similar state).
If you use this lora as a checkpoint for animeboysv3 on webui, you can make it somewhat more in character by setting the intensity of the lora to 1 or more, or by emphasizing the trigger word.
@ikaru8901 Hi, thanks for making your time to tell me! Yes, it has been resolved now! Tho making the weight more than 1 usually ends up make the picture distorted ><
@RaidenFan86
Regarding the issue of distortion, is it a matter of the prompt content, the total amount of tokens, the total weight of the other lora, or the GPU you are using?
I have not experienced any distortion in my environment. However, the reproduction of the eyebrow shape seems to be particularly tough. (The background is always wrong).
As you know, this lora and animeboysv3 do not work well together, so if you just want to make the character look like the original, it is better to use other checkpoint instead of animeboysv3.
I will grally submit some images created with animeboysv3 using this lora.
As a side note, I often use animeboysv3 with “aged down” or “cute” because it makes teenage characters look like they are in their 20s or 30s.
@ikaru8901 Yeah, my GPU are not that good, maybe that's why x_x
AnimeboysV3 is currently my favorite Checkpoint since it trained with male character, thus would always generate male character, unlike other Checkpoint than sometimes will generate females even when using Loras.
The thing is, it's quite hard to master since you need to put different tags than other Pony Checkpoint. It also need to put the tags in order so the result would be better. You can see how to use the proper tags for it in its pages.
But when you master it... It will make even better results than other checkpoint! XD
It can even make popular characters, and also some artist's artstyle without loras!
I always use it in my result, so you can see it in my account.
Maybe I should also put the result here afterwards~
@RaidenFan86
As for image distortion, this extension may help.
According to the article, when using the xl model in WebUI, there seems to be a bug in the intensity calculation that causes the highlighted prompts to produce abstract picturesque images when certain situations are encountered.
This extension would provide the ability to skip that buggy calculation.
https://github.com/hako-mikan/sd-webui-prevent-artifact