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
Repeats: 5
Epoch: 4
Image Count: 300
Network Rank: 32
Network Alpha: 16
Batch size: 2
Clip skip: 1
Training steps: 3000
Resolution: 768x1024
Train Resolution: 1024x1024
Model: ponyDiffusionV6XL_v6
Captioning: WD14 Captioning + Edit (SmilingWolf/wd-v1-4-moat-tagger-v2)
Trigger Words: raw, photo, realistic,
LoRA Weight: [0.8-1]
Without the tokens raw, photo, realistic, the model will generate 2.5D-3D and sometimes 2D styles, these tokens should be at the beginning after the score, if you can't get a realistic style, you're doing something wrong try to change the prompt
The preview should show that the first image is with Trigger Words and the second image is just with LoRA
Score, Trigger Words, Prompt
FAQ
Comments (12)
Photo 2
Without the tokens raw, photo, realistic, the model will generate 2.5D-3D and sometimes 2D styles, these tokens should be at the beginning after the score, if you can't get a realistic style, you're doing something wrong try to change the prompt
The preview should show that the first image is with Trigger Words and the second image is just with LoRA
Score, Trigger Words, Prompt
Where is my POG react. Just now decided to check in and it just came along. Hella based.
And it's better than I could have imagined. Bless.
So what is the difference between v1 and v2 of the models?
Difference in dataset and training
Smooth - v1 8 images v2 12 images
Anime - v1 8 images v2 100 images
Photo - v1 22 images v2 300 images
@prgfrg23 Thanks. So they both work with Pony! Great.
How did you even manage to make the photo2, it's so well done.
It's not so good with faces and more complex prompts sadly.
On the one hand it was easy, but on the other hand it was very difficult.
2 hours of searching the dataset, practically a week of editing each image and 3 hours of training.
100 images of women
100 images of men
100 images of other, animals, landscapes, food
All images were taken from Unsplash, which means there is no NSFW content in LoRA.
In v1 there were only 22 images, They were AI generated. No NSFW content there either.
There is no difference in SFW/NSFW content for LoRA.
LoRA can create any content, LoRA is completely model dependent.
@virtualfix6885 Well with NSFW content in Pony Diffusion V6 XL is much better, but the faces are not very good, faces are good only in portraits, the smaller the face is the worse
Thank you for those styles!
BTW did you see the <10Mb Nier Lora with Dim 1 Alpha 0.1? Maybe you can check if your styles would work with a smaller size too?
Anyway, thanks again 👍
I haven't tried reducing the size of the network with styles, but I have tried it with the character
200Mb Good.
100Mb Normal
50Mb Still ok but could be better.
Less than 50Mb I haven't tried.
The smaller the size the less LoRA will remember, with simple styles it might work, but with Photo Style I'm not sure.
Anyway, I'll try to do something about it.
I won't promise anything!
Details
Files
Photo 2 Style SDXL_LoRA_Pony Diffusion V6 XL.safetensors
Mirrors
Photo 2 Style SDXL_LoRA_Pony Diffusion V6 XL.safetensors
Photo 2 Style SDXL_LoRA_Pony Diffusion V6 XL.safetensors
Photo 2 Style SDXL_LoRA_Pony Diffusion V6 XL.safetensors
Photo_Style.safetensors
Photo 2 Style SDXL_LoRA_Pony Diffusion V6 XL.safetensors
Photo 2 Style SDXL_LoRA_Pony Diffusion V6 XL.safetensors
Photo_Style.safetensors
hf_model_48_18.safetensors
Photo_2_Style.safetensors
Photo2StylePonyXL.safetensors
SDXL1.0_LoRA_Photo2Style_Pony_6.safetensors
Photo 2 Style SDXL_LoRA_Pony Diffusion V6 XL.safetensors
Photo 2 Style SDXL_LoRA_Pony Diffusion V6 XL.safetensors
Photo 2 Style SDXL_LoRA_Pony Diffusion V6 XL.safetensors
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.