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
Flux dataset integrated as well as a smaller one for a few other miscellaneous things. Best version ever for sure IMO.
FAQ
Comments (18)
I put werewolf seggs in the V6 sample images but it got detected as "bestiality" by the vision model, have to wait for a mod to unblock it.
hmm lol they nuked it outright, i'll make a few more sample pics today then anyways I guess. Hope people like V6 regardless!
There, added a couple more.
Can I use this for my NSFW games / comics?
Uh, not quite sure what you mean, are you trying to like ask permission for something?
Use your eyes! You're not allowed to use the model in generation services like Discord bots or website generators, you're not allowed to sell the model, the rest is in the OPEN RAIL-M license and any work spawning from this model must also go under the same license, assuming the license say so.
I might update this again at some point, probably not soon though. Any feedback anyone has is always welcome though, of course.
Could you enable "Run on Civitai" permission of this model?
And could you make version that merged Hyper SD1.5 8 steps CFG for this model?
@LOL2024 Even if I did enable it it wouldn't go up, the generator checkpoints are limited to top 200 most popular or so, with slots only allotted to certain people. Plus it wouldn't be usable at the intended resolution here on Civit, results would look bad. I can probably put it back on TensorArt though, I had it there at one time but it got messed up somehow IIRC. As to Hyper I can look into it, no idea how well it'd work though.
@LOL2024 it's back on TensorArt now:
https://tensor.art/models/786372554702143690/Better-Pony-Diffusion-V6-For-SD-1.5-1
Hey just feeling lazy here , can you confirm that your model has a good basic anatomy understanding or not without using loads of quality prompts
It's about as good as it gets on SD 1.5 I think. Just be sure to read the whole page description, and keep in mind Pony V6 SD 1.5 was trained by AstraliteHeart with a 768px baseline resolution, and then further tuned at 1024px by me. It is not intended for use at the "normal" 512px SD 1.5 resolution range at all.
It seems like at least one person actually has trained a Lora on V6, so I'll note again in general that if you're gonna do so, do not train it at 512px, train it at 768 or 1024px. Particularly if you're using the CivitAI trainer and choosing this as a custom model you can (and should) just switch the resolution there in the box to 1024.
need more like this,, pony for sd
sometime the style is realistic,some time anime style,how can i control the style ?
Add "anime" or "photo of" in your prompt depending on what you want.
@Zenyth yeah, generally speaking 3d, cgi, realistic, photorealistic, raw, photo are the ones that can be put it pos or neg to control what they sound like, and then 2d, traditional media, illustration, painting \(medium\), watercolor \(medium\) are the ones that can be put in the pos or neg for sort of the opposite. Then of course also the actual Pony source_ tags have an impact as well
Note: I would have 100% replaced the baked VAE with Furception if it were possible, but the way he baked the WD one initially is unusual, you cannot just swap it out without getting bad results, it only works if you actually "Re-Encode Latent" to Furception in ComfyUI. I have no idea what he did to make this the case on an SD 1.5 model but yeah that's the reality, sadly.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.