02.09.2025
NAISDXL_PAINTERLINESS_V0.2
Recommended Settings:
- Sampling Method: Euler - DPM++ 2M - Heun
- Schedule Type: KL Optimal - Uniform
Prompt Recommendation:
masterpiece, best quality, absurdres, highres, safe, 1man, solo, mature male, smile, bara, singlet, nature, oil painting, traditional medium
masterpiece, best quality, absurdres, highres, safe, 1woman, solo, mature female, smile, curvy, sleeveless shirt, nature, oil painting, traditional medium
deformed iris, deformed pupils, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, deformed, bad anatomy, bad proportions, disfigured, fused fingers, too many fingers, bad hands, text, missing fingers, signature, watermark, low contrast, sepia
01.09.2025
ILSDXL_PAINTERLINESS_V0.2
Prompt Recommendation:
masterpiece, best quality, absurdres, highres, 1man, solo, mature male, smile, bara, singlet, nature, oil painting, traditional medium
masterpiece, best quality, absurdres, highres, 1woman, solo, mature female, smile, curvy, sleeveless shirt, nature, oil painting, traditional medium
Negative Prompt Recommendation:
nsfw, nudity, deformed iris, deformed pupils, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, deformed, bad anatomy, bad proportions, disfigured, fused fingers, too many fingers, bad hands, text, missing fingers, signature, watermark, low contrast, sepia
01.09.2025
PDSDXL_PAINTERLINESS_V0.2
Prompt Recommendation:
oil painting, traditional medium, score_9, score_8_up, score_7_up, rating_safe, 1boy, solo, smile, nature, canvas texture
oil painting, traditional medium, score_9, score_8_up, score_7_up, rating_safe, 1boy, solo, smile, nature, canvas texture, visible brushstrokes, loose brushwork, painterly, textural, artistic
Negative Prompt Recommendation:
score_6, score_5, score_4, score_3, deformed iris, deformed pupils, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, deformed, bad anatomy, bad proportions, disfigured, fused fingers, too many fingers, bad hands, text, missing fingers, signature, watermark
score_6, score_5, score_4, score_3, deformed iris, deformed pupils, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, deformed, bad anatomy, bad proportions, disfigured, fused fingers, too many fingers, bad hands, text, missing fingers, signature, watermark, anime, photorealistic, 35mm film
31.08.2025
PDSD1.5_PAINTERLINESS_V0.2
For photorealism enthusiasts, I recommend Cum_Miser's Cool Realism LORA. I've tried it and gotten excellent results. Because Lora was trained on Pony Diffusion v6 SD1.5, it's perfectly compatible with PDSD1.5_Painterliness.
Prompt Recommendation:
- oil painting, traditional medium, score_9, score_8_up, score_7_up, rating_safe, 1boy, solo, smile, nature, canvas texture
Negative Prompt Recommendation:
- score_6, score_5, score_4, score_3, deformed iris, deformed pupils, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, deformed, bad anatomy, bad proportions, disfigured, fused fingers, too many fingers, bad hands, text, missing fingers, signature, watermark
28.07.2025
(SDXL,PONY,ILLUSTRIUS,NOOBAI EP) OneTrainer fine-tune settings:
MODEL TAB
Weight Data Type - float16
Override UNET Data Type - float16
Override Text Encoder 1-2 Data Type - float16
Override VAE Data Type - float16
Output Data Type - float16
TRAINING TAB
Optimizer - LION
Learning Rate Scheduler - COSINE
Learning Rate - begin with 1e-05, if it is very baked sample try 1e-06. Determining the learning rate correctly is crucial. A lot of trial and error.
Learning Rate Warmup Steps - 0.25
Epochs - 10 (at least 500 dataset) - If you have smaller dataset, please consider 10.000 - 12.000 steps. If you have bigger dataset epochs are lower.
Batch Size - 1
Train Data Type - float16
Fallback Train Data Type - bfoat16
SAMPLING TAB
Sample after - 300 steps - PNG
BACKUP TAB
Save Every - 1 or 2 EPOCH
Notes:
Once you've prepared a good dataset, you can use these settings. You can prepare the concepts in your dataset with TAGGUI or OneTrainer Dataset Tools. If you have only one or two concepts, LORA is more suitable. For multiple concepts, I recommend full fine-tuning.
Minimum Hardware Recommendation for Fine-Tune:
SDXL - If you want to train text encoders as well, RTX 3090 can handle this. If you leave the text encoders as embeddings, you can do it with the RTX 3080. At least 16GB VRAM
SD1.5 - You have to train the text encoder anyway; 30 epochs is enough for small dataset. Bigger dataset lower epochs. The RTX 3070 can do this job. 8GB VRAM
For Lora Training at least RTX 3060 8GB for sdxl and sd1.5 (768x768 pixels dataset), only unet training with sdxl, no text encoders.
Please don't write that it takes too long with the RTX 3060, it is the minimum hardware you can train with.
22.07.2025
PAINTERLINESS NAISDXL
This model has been fine-tuned with ONETRAINER on NOOBAI EPSILON PRED v1.1 SDXL version.
Recommended Settings:
- Sampling Method: Euler - Heun
- Schedule Type: KL Optimal - Karras
- CFG Scale: 5 - 6
- Sampling Steps: 30 - 50
- Width Res: 864 - Height Res: 1152 (864x1152 - 3:4)
- Hires. Fix - Upsacaler: None or 4x Ultrasharp, Remacri - Denoising Strength: 0.3 - 0.4
There are 3 version BF16, FP16 and FP8. You can use FP8, if you have very limited GPU RAM.
Prompt Recommendation:
- masterpiece, best quality, absurdres, highres, safe, 1woman, mature female, curvy, oil painting, traditional medium
- masterpiece, best quality, absurdres, highres, newest, 1girl, solo, curvy, perfect feminine (muscular:0.6) body, buzz cut, oil painting, traditional medium
Negative Prompt Recommendation:
- deformed iris, deformed pupils, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, deformed, bad anatomy, bad proportions, disfigured, fused fingers, too many fingers, bad hands, text, missing fingers, signature, watermark, username, low contrast
- nsfw, nudity, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, sepia, ugly
- nsfw, nudity, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, blurry, ugly, sepia, deformed iris, deformed pupils, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, deformed, bad anatomy, bad proportions, disfigured, fused fingers, too many fingers, low contrast
21.07.2025
PAINTERLINESS ILSDXL
This model has been fine-tuned with ONETRAINER on ILLUSTRIOUS v2.0 SDXL version.
Recommended Settings:
- Sampling Method: DPM++ 2M SDE - DPM++ SDE
- Schedule Type: Karras - KL Optimal
- CFG Scale: 5 - 6
- Sampling Steps: 30 - 50
- Width Res: 864 - Height Res: 1152 (864x1152 - 3:4)
- Hires. Fix - Upsacaler: None or 4x Ultrasharp, Remacri - Denoising Strength: 0.3 - 0.4
There are 3 version BF16, FP16 and FP8. You can use FP8, if you have very limited GPU RAM.
Prompt Recommendation:
- masterpiece, best quality, absurdres, highres, safe, 1woman, mature female, curvy, oil painting, traditional medium
- masterpiece, best quality, absurdres, highres, newest, 1girl, solo, curvy, perfect feminine (muscular:0.6) body, buzz cut, oil painting, traditional medium
Negative Prompt Recommendation:
- deformed iris, deformed pupils, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, deformed, bad anatomy, bad proportions, disfigured, fused fingers, too many fingers, bad hands, text, missing fingers, signature, watermark, username, low contrast
- nsfw, nudity, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, sepia, ugly
- nsfw, nudity, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, blurry, ugly, sepia, deformed iris, deformed pupils, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, deformed, bad anatomy, bad proportions, disfigured, fused fingers, too many fingers, low contrast
20.07.2025
PAINTERLINESS PDSDXL
This model has been fine-tuned with ONETRAINER on PONY DIFFUSION V6.0 SDXL version.
Recommended Settings:
- Sampling Method: DPM++ 2M SDE - DPM++ SDE
- Schedule Type: Karras - KL Optimal
- CFG Scale: 5 - 7
- Sampling Steps: 30 - 50
- Width Res: 864 - Height Res: 1152 (864x1152 - 3:4)
- Hires. Fix - Upsacaler: None or 4x Ultrasharp, Remacri - Denoising Strength: 0.3 - 0.4
There are 3 version BF16, FP16 and FP8. You can use FP8, if you have very limited GPU RAM.
Prompt Recommendation:
- score_9, score_8_up, score_7_up, rating_safe, 1woman, mature female, curvy body, oil painting, traditional medium
Negative Prompt Recommendation:
- score_6, score_5, score_4, score_3, deformed iris, deformed pupils, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, deformed, bad anatomy, bad proportions, disfigured, fused fingers, too many fingers, bad hands, text, missing fingers, signature, watermark, username, low contrast
19.07.2025
PAINTERLINESS PDSD1.5
This model has been fine-tuned with ONETRAINER on PONY DIFFUSION V6.0 SD1.5 version.
Recommended Settings:
- Clip Skip 2
- Sampling Method: DPM++ 2M SDE - DPM++ SDE
- Schedule Type: Karras - KL Optimal
- CFG Scale: 5 - 7
- Sampling Steps: 30 - 50
- Width Res: 768 - Height Res: 1024 (768x1024 - 3:4)
- Hires. Fix - Upsacaler: None or 4x Ultrasharp, Remacri - Denoising Strength: 0.3 - 0.4
There are 3 version BF16, FP16 and FP8. You can use FP8, if you have very limited GPU RAM.
Prompt Recommendation:
- score_9, score_8_up, score_7_up, rating_safe, 1woman, mature female, curvy body, oil painting, traditional medium
Negative Prompt Recommendation:
- score_6, score_5, score_4, score_3, deformed iris, deformed pupils, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, deformed, bad anatomy, bad proportions, disfigured, fused fingers, too many fingers, bad hands, text, missing fingers, signature, watermark, username, low contrast
Explanations:
score_9, score_8_up, score_7_up: It is rating system for Pony Diffusion. If you don't use it, you will get crappy results.
oil painting, traditional medium: If you want oil painting on canvas effect, use it.
Description
It is in ALPHA stage
I will train and fine-tune more
V0.1 - V0.5 : ALPHA
V0.6 - V0.9 : BETA
V1.0 - planned to be all my training data depleted
FAQ
Comments (25)
why not lora?
loras may be one or two concept dilemma, in my case full fine-tune is better. I can make lora from these models, I will think about it.
Holy shit, dude, I think this might be my favorite model in a very, very long time. Mainly because you've managed to do something so many other models could not: you made it so that characters don't have the same face every damn time, with an impressive level of prompt adaptability and adherence. I hope you continue your work on this, it's something special in a sea of models that all look the same.
Thank you so much. You've summed up exactly what I want to do. I'll be adding more concepts in the future.
I was wondering why all the women were so thick.
Maybe I like thick women. :)
This model has some difficulty in generating horror-themed images ... but when it does get something good, my goodness is it good. I need to tweak how I word the prompt to figure out how to get better results, but I can absolutely see this being a great horror generator ala Francisco Goya
Thanks for your feedback, I'll work on it. I'll especially try to emulate what Goya did during his time of isolation.
Wow, SD1Pony actually looks sharp. And "anatomy-gacha" has better success rate.
Yeap, you are right :))
Sergun Will you consider adding hand gestures into your dataset?(SD1 really struggles with them) It would be cool to be able to do thumbs up, thumbs down, heart hands, middle finger etc. as reaction pictures.
Do you plan, or already added some artist styles?
Cum_Miser Good idea, I need to create a dataset for hand gestures. This will be available in future versions.
Cum_Miser Some characters and artist styles are missing in sd1pony, or they disappeared when I trained it with the new dataset.
Sergun Great, to hear. Maybe gestures also improve hands in general. (Maybe feet also haha?👀)
But, do have plans to add any artists styles?
I've also been tinkering with Onetrainer, trying to train style-loras for MilfShaper2.3 and Pony6SD1 base model. Performance hit wasn't worth slight aesthetic improvements.
But now, training on Painterliness, results are much more core coherent, at lest for upper body.
Cum_Miser When you train a model, you don't know what the outcome will be. So, I trained four different models with the same dataset. Each one took different elements of the dataset. It would be great if I could combine them all, but that's not possible.
Cum_Miser I used the Lion Optimizer with the Cosine Schedule on OneTrainer. This gave me the best results. It was a lot of trial and error.
Cum_Miser If you want to train Lora, you can do it very easily with OneTrainer. Determining the learning rate correctly is crucial; you can train Lora with one or two concepts using 3e-5 UNET training.
Sergun Not sure if it's answers my questions but thank you for reply.
Cum_Miser I am planing to painting styles first, after digital art styles and underground styles.
Sergun You mean general styles (merge of similar artists), or separate artists (like you did in MilfShaper2.3)?
Cum_Miser all of them :))
Sergun Hell yeah!
Hey thanks man... mind sharing the Onetrainer settings for the finetune? I didnt had much success fine tuning with OT SDXL yet...
Of course, man. I will share the OneTrainer settings in the description section.
Sergun Thanks! much appreciated. I tried several finetunes with a fairly decent datasets using ADAMW, Adafactor and ADAMW8b and tweaks... like loss function, batch sizes, Train teenc, etc..
and always... either doesnt learn at all or burns into crazy noise and garble suddenly.
I ll try your settings.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.