Not Artists Styles for Pony Diffusion V6 XL

Not Artists Styles for Pony Diffusion V6 XL - Smooth Anime 2 [M]

NSFW

Styles are trained on synthetic data and do not copy the styles of artists!

Photo 2 - is trained on real data (Dataset from Unsplash)

[M] Merged LoRAs

Recommended LoRA Strength (Weight): 0.8-1

Preview generated without using ADetailer & Hires. fix

Description

Smooth 2 + Anime 2 = Smooth Anime 2

FAQ

Comments (14)

MalessarFeb 9, 2024

CivitAI

a 2.5d or semi realistic style would be great

obuyb404

Author

Feb 9, 2024

There is already a 2.5D style for the Pony Diffusion V6 XL

https://civitai.com/models/297619?modelVersionId=334933

MalessarFeb 9, 2024

@prgfrg23 it is so. will you have another go at photo style? =) the quality of your styles is awesome which is why i wa ssuggesting it haha

obuyb404

Author

Feb 9, 2024· 1 reaction

@Malessar I'm glad people like my models and I'm even a little shocked at how many times they've been downloaded 14K times! 0_0

I'm working on improving Photo Style

The hardest part of model training is finding a good dataset.

So I have to generate or use images already created by someone else in the same style.

At the moment all my models are trained on AI generated images.

97824Feb 11, 2024

@prgfrg23 I found that if you merge pony with some compatible realism model you can img2img gens to slowly get a real look that is faithful to the original. But without merging checkpoints I found it was impossible. I forget what I used but it was some recent generic porn realism model that seemed to work best (or maybe realvis3.0 I cant remember). If you look at the 2.5d or photo realistic attempts at pony now, they basically just completely change generation, any way just thought I'd share that random thought.

2198848Feb 9, 2024· 12 reactions

CivitAI

photo sytle needs work

obuyb404

Author

Feb 9, 2024· 5 reactions

Well, you could say I'm working on it.

The biggest problem is that I need a dataset of at least 100+ images.

When I created the dataset for Anime 2, I thought I'd go crazy.

Also the Photo model was kind of an experiment, I didn't even think it would work.

2198848Feb 10, 2024

@prgfrg23 why not just use stock photos or film stills

obuyb404

Author

Feb 10, 2024· 1 reaction

@julianarestrepo I don't know. I'll figure out the dataset anyway. =)

heathergreen95Feb 12, 2024· 1 reaction

CivitAI

Edit: working now ~~These LoRAs do not work in ComfyUI.~~

obuyb404

Author

Feb 13, 2024· 2 reactions

CLIP Set Last Layer -1 or -2 ?

-1 gray picture with noise and spots

-2 picture looks the same as in the preview

If you do not add the CLIP Set Last Layer node, the result will be exactly the same as -2.

If anything I apologize I hardly ever use ComfyUI so I don't know how it all works there

TwinklingSquidFeb 14, 2024· 1 reaction

I'm seconding prg's comment!

They most certainly work on Comfy. Make sure your clip skip is on -2 and also make sure you read through the description on the Checkpoint post thoroughly. There were a lot of tags I was originally missing that made it seem like nothing I did worked.

My other suggestion is to couple these with a more-detailXL or dream-artXL Lora. I got much better results from these style loras when combined with one of those.

UO_AI_ResearchFeb 19, 2024

My experience is that these LoRas work just like any other Lora with SDXL in comfyui. Only comment is that they are a bit weaker than other LoRas so use a strength near 1.

heathergreen95Feb 24, 2024· 1 reaction

@prgfrg23 CLIP -2 fixed the issue for me. Thank you, and I apologize for this late response. I will edit the original comment

LORA

Pony

by obuyb404

Download (Beta) View on CivitAI