Pro Version of Realism Pony V4-V5-V6 now Available on My Patreon
Onsite generations are permanently available on these models:
š Realism_By_Stable_Yogi V3: https://civarchive.com/models/166609?modelVersionId=992946
To keep this model available for Image generation on Civitai, please place your bids here and Iāll make it available for onsite generation:
https://civarchive.com/auctions/featured-checkpoints
Join me on Patreon for exclusive perks and early access to unique resources.
To discuss custom LoRa's or models, feel free to connect on Discord.
š Like this model to keep me motivated and inspired to create more!
š¬ Drop a comment and let me know what you'd love to see next.
š Review this model to help me improve and make even better creations.
š Hit that notification bell to stay updated with my latest models and updates!
Important Usage Tips
Add Stable_Yogis_PDXL_Positives at the beginning of your prompt section.
Add Stable_Yogis_PDXL_Negatives-neg at the beginning of your negative prompt section.
Description
FAQ
Comments (28)
you sir are a genius, keep it up!
Thanks !
I loved this checkpoint before realizing you had custom embeddings; now, I want to marry it.
Generates great looking characters using CLIP Skip 2
Thanks for the feedback.
thanks for this one!
Thank you for the feedback. š
Just wow. Unbelievable good.
Thanks for the feedback. š
nice
I trained a personal lora character with Civitai's onsite trainer and the training preview images are incredibly good, but when i use the lora in Comfy, the results are deeply washed out. This is using the default workspace + lora loader and suggested settings. Generating without the lora produces awesome images.. so what is the mismatch between my lora and the model ? Is it a Comfy bug ?
You can use Realism_V3_VAE. Use Forge or A1111āComfy is a mess as always. All my images are generated in bulkā1k, 2k, 5kāand all these are testing images created before launching any model. Later, I release all these images.
So, I never face any issues with any LoRA or training, etc. (š Work with a packed brain like A1111 or Forge, while Comfy feels like a brain explosionāwires hanging everywhere, spending a day fixing one workflow to just generate a few images. Such a waste of time.)
@Stable_YogiĀ I trained on Realism V3 Lightning, I guess this was a bad idea ? I will retrain my character on V3 VAE, but I need tips about the settings.. i usually leave all tags by default, then 20 epochs and unet rate 0.00100, clipskip2, and Prodigy, but these might be recommended for other models, what about this one ? which settings should be set on Civitai's onsite trainer, because theres a lot of room for error there... my dataset is 35 good quality pictures from a photoshoot.
@randomchatter1234776Ā
Train with Realism V3 VAE
If a woman, then remove 1girl, 1woman, and add a unique name for her with numbers and symbols like 456freddy@76
unet rate default
clipskip1,
and Prodigy,
dataset with 35 good quality pictures is fine.
Set 5 repeats
20 epochs
i never trained on civit trainer so i am not sure if that like Kohya or a different UI.
try these and you will get closer.
@Stable_YogiĀ @Stable_YogiĀ just realised the lightning trained lora i did earlier needs to be run at 125 steps to get a decent image, it might be due to me applying wrong training settings. but i will retrain it on V3VAE instead, following your instructions. If this works out great you will be rewarded. t.y.
@Stable_YogiĀ I am getting incredibly good results following your training settings advice. My war against Flux is now over. I deleted everything Flux related and nearly every other XL checkpoints and loras i had because this replaces them all. The only thing missing is good "cum'', it looks rather blocky or grainy, any quality suggestion for that ?
@randomchatter1234776Ā I feel blessed i ended a war š
use weight in captions like (cum:1.3) Or use a lora.
If dosnt work let me know i will train a lora and share it with you.
I hope you stay peaceful now.š
@Stable_YogiĀ Are you referring to this? SG161222/Realistic_Vision_V3.0_VAE
I'm not finding the Realism_V3_VAE on google.
@Kierkegaard420Ā Ref to this https://civitai.com/models/166609?modelVersionId=992946
@Stable_YogiĀ Thank you! :)
V5 XL seems quite good . improvements from V4 are very tangible and among those improvements the most I want to praise is a) better anatomy and prompt adherences(checkpoint listens to your prompt , trying not to leave anything behind , and does her best to do your bidding ) , b) lighting , sharp crisp focus , glossy texture , glittering effects , all work very well with deep contrast . I personally feel V5 is definitely superior to earlier versions !
Thanks for the detailed Feedback. I appreciate your input. Thanks
@Stable_YogiĀ your model can do or add what other can't . in that light , quite unique so I haven't yet get bored using it because it delivers something new everyday I have missed yesterday .
@ttssmm27806Ā Thank you for the love and experiments, you made my day !ā¤
Fantastic model. The bodies it produces, jhheeez, u haven't just sourced any old bodies, you found the best, great diversity in women, particularly with regard to traditionally underrepresented ethnicities, bold, striking images which purr with beauty when generated, I don't usually sit to watch wallpaper dry, or the kettle boil, but I will gladly sit 2.5 minutes staring at the progress bar in eager anticipation as I just know this model will craft an image of delight (or at least 80% of the time in a decent size batch lol). Great work Yogi - don't stop what you're doing. Namaste
Wow, thank you for such an amazing review! Glad you're enjoying the modelāI'll keep striving to bring more magic. Namaste!
for sure, Indian women weren't so beautiful in any other model, but this one is AWESOME !
TOTALLY AGREE! Great Work!