SnowPony
Best for 2D / Anime style with better backgrounds.
Alt version: more anime, better colors, textured sketch style
Note: Forge seems to generate a different model hash breaking crossposting, not sure why it happens.
Boo! 👻
Description
FAQ
Comments (37)
Missing VAE, no problem but still needed info.
Since release of Pony i search a Model wich is better with Background/Lighting, will see if this is it.
Thanks for the heads up!
I keep forgetting to bake the vae. It takes a long time to upload, sorry.
Thanks for this comment I couldnt figure out why my images were coming out grayscale!
@r_ain What VAE do I need to use? I tried every one I have but I get desaturated images with white spots on them no matter what.
@billtheduck0001 I use this one, otherwise I can bake the vae in and upload in the next 1-2 days or so https://huggingface.co/madebyollin/sdxl-vae-fp16-fix
@billtheduck0001 Added baked VAE version.
[AWARDED] open bounty for image creation to test backgrounds https://civitai.com/bounties/3137
I think it hits a very nice sweet spot between style flexibility and image quality. Definitely a keeper.
I tried RainPony before and liked it too but compared to that model this one doesn't have the semi-realistic style baked in quite as strongly which means it plays nicer with loras and it's easier to prompt a variety of styles. Also the backgrounds seem to be even higher quality in many cases.
Thanks for the review, it is what I was going for with both models. I like the images you've posted.
I like this style but I would like to merge it with another model to get more anime features for example I don't want a realistic nose. For some reason I cannot merge this model do you have a solution for this problem?
Should be easy to merge, same as any other model. What are you trying?
I tried merging 3x3mixXLtypeF and your model. It just says "Error merging checkpoints: 'snowpony_v10.safetensors'". On console it says something among the lines if theta func2 else None KeyError ´snowpony´. I've merged with 3x3 before without issues. Thanks for replying! =)
Got it working after running it with No interpolation once
Hello is it possible for this model to appear on civitai generator?
I love this checkpoint and is my usual go-to. But I've noticed that skin tones that aren't white usually fail and that's a pretty big issue for a character creator like me. While I don't consider it a huge problem compared to the pros of the checkpoint itself I wanted to ask if a v2 where this is fixed (or atleast improved) could be possible. Keep up the amazing work you're doing because I love it :)
Thanks for your review. Yes you are right, I will consider it in the next version. It won't help you much as a character creator, but for generations I usually try the below:
Prompt some tags "tan", "brown skin", "dark skin", "very dark skin"
Use Vectorscope to lower brightness
Skin tone Slider Loras (or similar)
Thanks for the tips! I will try to apply them
Amazing work. This is the best checkpoint that I've used on PonyXL by far (and I've tried over 50 at this point). The only issue is sometimes I get asymmetrical eyes EVEN AFTER I face detailer-ed the original output... Inpainting twice isn't really enjoyable. Any idea why that happens? The prompt "symmetrical eyes" does not really help...
Thanks! For the eyes, are you inpainting the whole face? What about trying looking up, down, away? Example prompt and seed?
@r_ain I think it's something to do with the character LORAs I use, when its a generic prompt, the output is okay... thanks for responding though :)
The finished image turns out gray with white dots on. Does anybody know how to fix this?
You are missing VAE, you can add it or use the bigger file labelled full model on this page (6.62 GB)
@r_ain I only see one file on this page (6.46GB) that doesn't have VAE. Am I doing something wrong?
@dasefa2395 it is in the files dropdown on the side, underneath hash and AIR
@r_ain I think it would be a good idea to put the full model as the primary download option, since that dropdown isn't that intuitive for most people
I used the Full file and loaded it as a checkpoint but still have the grey with white spots. Anyone have any ideas what's wrong?
Dude, i got so many great results with this checkpoint. I've been using VXP almost all the time, but this one is much more compatible with the character's Loras that i've been using. Great for Anime style.
thanks
I've extensively tried several dozen Pony models over the last 4 months, and since the middle of August when I first found this, it is now my default go-to when testing new character loras. Prior to this I was using ONE FOR ALL, but that checkpoint seems to fall apart when too many prompts are put in, has a hard time with faces when the face isn't the focus, even with adetailer, and cannot handle background environments that well. While I sometimes have face generation issues still even with this checkpoint, I'm assuming that's just the nature of Pony in general, and it not liking certain seeds.
I can't really say what could be improved. This seems to get very original artstyle accurate with most character loras, with some subtle improvements, which is exactly what I'm looking for.
Thanks for your review
Details
Files
Available On (3 platforms)
Same model published on other platforms. May have additional downloads or version variants.