This is the SD 1.5 version of Pony V6 fine-tuned at native 1024px on ~5000 hand-selected images (some of them borrowed from my Zootvision model's datasets). Each image was captioned both with Florence-2 Large "More Detailed Mode" rich captions and also Booru tags from WD-VIT-V3. Use this the same way you'd use Pony V6 SD 1.5 normally, and just uh, enjoy the pretty objectively better overall aesthetics.
Important notes:
in A111, use Clip Skip 1 (NOT 2) and in Comfy just do not use the "Clip Set Last Layer" at all, with this
DO NOT use any VAE besides the one that is baked into the checkpoint, it will fry the image
You will probably get worse results from weird "in between" resolutions like 720x1280 than you will from standard ones like 768x1024 and 832x1216
Basic positive prompt: score_9, source_whatever, rating_whatever, your tags or natural language description here.
Basic negative prompt: score_3_up, score_4_up, score_5_up, sketch, (simple background:1.2).
Recommended steps / sampler CFG: typically Euler Ancestral at CFG 7.0 with around 25 - 35 steps is a good starting place. The DPM++ SDE GPU family of samplers can also be good with this at lower CFG (4.0 - 5.0) if you're going for realism in particular.
Generating at 512x512 is NOT recommended, as this model was originally trained by AstraliteHeart at 768px, and my additional training was entirely done at 1024px.
Description
Initial Version.
FAQ
Comments (8)
did you remove the need for "score_" garbage that makes the model annoying to use?
i cant seem to understand in what reality a person would ever want to generate anything other than the highest possible score image...even in the cases of using it for negatives....if you are training in low score images just caption them "ugly, distorted, gross" and be done with it....the whole score system has to be one of the most moronic things to train into a image gen model.
after making about 40 images, the score system seems to still be a requirement. however, the name of the model is true...this is better than what pony gave us for 1.5. very good job on penises and male figures, thank you...most models forget that you have to be able to gen BOTH sexes to make really good nsfw images. has pretty good prompt adherence too. good job, hopefully youll consider what i said about no one will ever purposely gen anything less than highest score and remove that stupid tag from the training.
also thank you for keeping it in clip skip 1! people dont realize that when you make a model skip 2 it cant be merged successfully with a skip 1 model without extreme malformations....
the model is absolutely fantastic for mixing.
@omegablast20023899 To be clear this isn't a "mix", as I said in the description I literally tuned Base Pony SD 1.5 on images I selected and captioned myself.
As far as the score thing, I did not use the score system myself when captioning (as also said in the description, just Florence 2 Large detailed caption + WD-VIT-V3 booru tags) however, it's still relevant of course for all of the massive amount of data that was in the model to begin with. So I'd say basically you'll get generally better looking results regardless, but overall score_9 is still beneficial.
Note that the 1.5 version of Pony never needed the whole list of scores to begin with, unlike the XL version, it works with just score_9.
Also Pony SD 1.5 was always Clip Skip 1, BTW.
Glad you're enjoying this version, in any case!
@diffusionfanatic1173 im aware that its not a mix. i was simply stating that it is fantastic FOR MIXING. not that it is a mix. look at my posted pictures after i mixed it in with some of my own models for results.
it brought color and lighting to the next level for me in my model. and coherence!
i posted about 10 but most are stuck in "analysis" limbo....thanks civitai!
@diffusionfanatic1173 was it clip skip one? because with it i was getting pure garbled mess gens. anyways who cares now :) your version of that is hands down superior.
@omegablast20023899 If you use Comfy, "Clip Set Last Layer = -1" is NOT the same thing as just straight up not using the "Clip Set Last Layer" node, so maybe that was your problem with the original version. Anyways, as I said in the description with this if you do use Comfy, just don't use that node in this case, you don't need it.
I'll probably be uploading a V2 of this a bit later today, trained on about 200 more images I picked out to fill in what I felt were a few more gaps. New Zootvision ( which is still my main focus and IMO definitely a better model by a fair amount in all honesty ;) ) soon too!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.