If you like my work, drop a 5 review and hit the heart icon. it's free and keeps me motivated
All my models are officially hosted and maintained by me on Tensor.art . use my Exclusive and public model for free on tensor.art
v5.0 civitai onsite generation enabled : 04/01/25
v5.0 final release : 01/01/25
v5.0 pre relelase update : 25/12/24
Check tensor.art for V1,V2,V3 online generation support
You can run Prefect Pony XL and use its API on SinkIn
currently, only v5 and v4 are available for civitai generation.
v3,v2,v1 are retired from civitai generation. check other version details for alternative on site generation service.
WIP: Realistic
V5 Update :
onsite generation : https://tensor.art/models/815661199413897287
pre-release : https://tensor.art/models/810892802167877454
Added trained Lora
Cleaned style
Baked in VAE
Extra trained data of 15000 steps
Better Lora support.
V4 Update :
onsite generation : https://tensor.art/models/794513755405953653/Prefect-Pony-XL-v4
Added trained Lora
Cleaned style
Baked in VAE
Extra trained data of 15000 steps
Better Lora support.
V3 Update :
onsite generation https://tensor.art/models/772139385270444718
added some style to it
should give better face and eyes
better NSFW supports
wide range of Lora support
V1 lightning :
8-steps
DPM++ 2M
V2 update:
onsite generation https://tensor.art/models/754887012743423435
has less style influence
can work with wide range of loras
better anatomy
better color
V1 update:
On site generation closed for v1 on civitai for limitation of civitai.
try tensor.art if you want to use the online generation for v1
https://tensor.art/models/724940954567461228/Prefect-Pony-XL-v1.0
If you like my work then drop a 5 review and hit the heart icon. it's free and keeps me motivated
Suggested settings:
I had CLIP skip 2 on every image
I had ENSD: 31337 all of them
All of them had highres.fix or img2img at higher resolution.
I don't use restore faces
I use afterdetailer for face details
Tiled diffusion for img2img upscaling and sometimes Noise Inversion for some more extra details.
4x-Ultrasharp upscaler
NOTE: if you find any prompt of the preview images look familiar it's because I've taken them from other model review images, Credits to the original authors. Thanks for the benchmark.
Commission open in Patreon.
Get early access and Exclusive NSFW Lora in my Patreon .
Support my work by joining any one of them and get early access to all my upcoming loras and other perks such as fan requests and Discord role.
Join my Discord Server
Description
V3 Update :
added some style to it
should give better face and eyes
better NSFW supports
wide range of Lora support
FAQ
Comments (143)
And what does V3 do?
V3 Update :
added some style to it
should give better face and eyes
better NSFW supports
wide range of Lora support
Very good model, only bad thing its I feel its that it make the style loras worse
Me finally feeling comfortable w/ the checkpoints style and solving difficulties.
Goofy_Ai: Sorry, baby. New version. Git good again.
How is it basically different from Pony?
good model, has better detail than AutismMix but less expressive/more stylized....
Yes.
Looks nice for anime portrait.
It doesn't follow my prompt at all, help?
what prompt are you using and what output are you getting?
@Goofy_Ai It's weird, sometimes it works sometimes it doesn't. I tried this basic prompt and it worked.
Prompt: score_9, score_8_up, score_7_up, score_6_up, 1girl, mature, older, indoors, kneeling, carpet, v arms, looking at viewer, long hair, casual clothes
Neg: score_4, score_5, score_3, score_2, score_1, source_pony, source_furry, hyperrealistic, realistic, 3d, monochrome, text, watermark, logo,
But as soon as I alter it to be any more nsfw, it creates completely unrelated images with simple backgrounds and noisy jagged lines
@fredakan3047164 i tested it with multiple character/nsfw pose/random prompts and never faced any issue . at least not this much. which UI are you using?
@Goofy_Ai Forge
@fredakan3047164 gave you the prompt fix. check inbox
Very good Pony XL checkpoint ! I love it <3
The best anime base model
3.0 hands and eyes have very good performance
I have based all lora training on this model
Makes women out of men :/
Tags and negative tags are not helpful. On v2, the problem is not as noticeable.
noted
I noticed that sometimes score tags or quality phrases make ignore the hints
@clnd I don't use any quality tags other than score 9-6. I never realized which tags are trig, but on version 2 and 3 I had a woman just popping up out of nowhere with the same prompt.... Or once it was a pumpkin with a penis. No kidding.
@LapaKotika once I created a simple background without people and the word anime always added a girl to the image, this was on version 2
you also can't get a simple cat image, the model will always draw a anime-girl-cat, but if you add clip attention multiply and play with the settings it will get cat (I use comfy ui)
@clnd i tested split-level shot and with prompt: score_9, score_8_up, score_7_up, lake, half underwater
Neg: 1girl, girl
50/50 image turned out to be a girl. Eeeeehhhh. Really annoying.
The hair color is inconsistent and usually made darker than other models
mentioning anime style (source_anime) can add color
@clnd already did it tho...
Comparing to v2, v3 is more leaning to western style, need `source_anime` in positive or character lora if you want waifus.
Seems paying more attention to detailed tags in prompts, generating complicated outfits much better. Easy to spot that images felt more "contentful" in V3.
The background is detailed and magnificent as always. A more Prefect-styled upgrade of v2, which I think is not a bad direction.
One more point I'd like to praise. Generating bushes or forests in V2 usually result in Photoshop paintbrush-like repetitive pattern; that is also fixed in V3. This model is more powerful and all-round than ever.
I feel that the brushstroke style in V3 is much heavier and thicker compared to V2 and V1. In fact, each generation becomes progressively heavier. Personally, I don't quite like this style as it makes the image feel less clean, and the style starts to resemble American comics. V1 is the cleanest, but its content is not rich enough, which is why V2 is my favorite.
Great model!
what vae do you recommend with this?
no need. vae baked in.
newest version dosent really work with female furries that well, it forces alot of human parts onto them
human ears, noses, face structures, ect, and dosent really listen to blacklisted prompts for female furries
furries were never been the main focus.
@Goofy_Ai tho it does work very well with males furries tho, so idk, maybe some more training on it to make it not ignore negative prompts may be good?
@PyroAsrielCAD will have to add some furry training in future versions
@Goofy_Ai that would be great! maybe tag a few of the facial features for humans so they can be easily disabled with negative prompts
V3 works very well for me. It gives more contrast in color and is always wet at first, but adjusting the CFG scale slightly and adding 'wet, wet skin' to the negative prompt solved the issue.
And as always, it also generates good hands. Thank you for making this!!
need V3 cleaned style.
I have a little issue with prompts like pig-tails or bunny ears. Lots of unwanted pigs and bunnies. Any tips on how to prevent that?
try adding negs like "animal or bunny" and or using "playboy bunny" for positive.
@skullx198x980 Tried that, no luck. Stupid question, but does attention/emphasis work in negative prompts? In other words, if i put ((animal)) tag in negative prompt, is it less likely to appear?
@humanb331ng572 I sometimes put higher weight into negative prompts and it is a coinflip whether it helps or not. It's always worth a try though. Go with 1.4 at least
can't wait to give v3 a test run. comeon! download faster!
goof! what hrfix settings do you use?(aside from 4xultra, like steps and dn etc..)
dank af yo
很棒的模型！
The generated character has many layers of flesh on the side waist. Is there any tag to avoid this?
I have never faced this issue or seen anyone facing it. hard do decode
happens to me since V2 too. I mainly generate in 1080x1920 with HiDiffusion. Putting Deformed in negative or removing Skindentation form positive ain't working. It looks like 2~3 too obvious ribs.
Nice model
Anyone have suggestions regarding the best sampler to use?
@Eminic You can't go wrong with Euler but i mostly use DPM++ 2M Karras with good results.
nice
Some decent improvements over v2 in v3, but a bit too much stylistic bias.
I like the new background details in V3, but the characters are too difficult to tame
I wonder if the checkpoint name was misspelled or what? I'm a bit confused actually but yea it worth to be called "Perfect Pony" :D
Nice
Can I sell images created using Wonderful Model and Lora?
I'm drawing for social media and my posts have gotten way more popular ever since I am using this model.
It's not perfect, but I am simply redrawing the bad parts in Krita and refine with lower % which blends things nicely. 10/10 recommend.
good
You haven’t indicated settings for the model: cfg, steps, sampler, etc
For buz
fire
cool
вцывыв
as the borat would say. VERY NICE
cool model, very nice
Prefect is absolutly Perfect !
I use mostly the v2 cleaned style, well done ! Love it <3
nice
Not the best place, but what is currently the best GUI to use for image generation? I tried to run this in ComfyUI, but it just crashes for me.
Is AUTOMATIC still the go to?
i use autoamtic1111 for all my testing.
i use this as my main and i only use comfyui, so im not sure why yours is crashing
@llloydbanker912 idk. same i use it daily and its up on different site for generation . all working.
Maybe my GPU is too weak? RTX 3060 Ti
But I will try and look into running ComfyUI
is it possible you share a workflow used to create example images?
@bcn sorry imnlwte to the response I'll post a pic but I only use 3 nodes. "Enhanced lorra loader" "Enhanced checkpoint" and then one more node to save results
Your model is awesome, but unstable for Loras' art style. Sometimes, i get the Lora rendering but not the full body.
for styles i suggest using v2
v3 is mostly for characters
great model, it's been a while since I generated images, I totally forgot how a pain sometimes is to get the background not looking weird
this model does not work with ControlNet?
works
@Goofy_Ai ControlNet works for me with other models, but specifically with the Prefect Pony XL model it refuses to work and gives the following error: RuntimeError: mat1 and mat2 shapes cannot be multiplied (154x2048 and 768x320)
I've been looking for a solution to the problem for 4 hours already😢
@crashpro8194 which controlnet model are you using?
@Goofy_Ai i use stable-diffusion-webui-forge
there is ControlNet Integrated
@crashpro8194 Im asking for the controlnet model. Because for me controlnet is working
@Goofy_Ai @Goofy_Ai What kind of controlnet model do you use? Can you provide me with the link or name? Please.
@ILIiliiIIIllIILliIl i dont use any controlnet.
I LIKE THIS
Cool
Its just perfect! ♡ Thank you c:
love this!
nice one
nice
this actually more fine tuned than autism in all aspects
i think to
Great!
amazing!! thanks
Nice model, but sometimes I'll get these random splotches of colour on an image. No idea what causes that.
cool
First model I've ever used. Won't be using anything else any time soon.
Excellent model, but i have one question,This might sound dumb but...sometimes it generate extremely realistic picture and i have no idea why...any clue?
add 3d/cgi/realistic to negative prompts
@nekocode368 ty
Somtimes hands mess up
sugoi desu yo but it really hates penises
Let us use v1 again!
You can use it on tensor art
v3 of this model is incredible, I actually think it is the best pony derivative model that has ever been created in terms of quality
Very good with the backgrounds and realistic loras
8/10
Love it
I'm actually amazed by V3. I can't believe how well it does hands.
Also body proportions are perfection. Anime faces with sem-realistic style super easy to achieve.
Finally have my Autismmix replacement. 10/10. seriously excellent work
bro is kinda professional in AI models
very high tolerant on using multiple LoRAs known to break image. I am very very impressed. and now I need to dl it with my little space left!
It also gives amazing results with no LoRAs at all! Better than any other checkpoint IMO
Some somehow managed to never have used this model until today, and boy have I been missing out. This is phenomenal!
it's been my favorite for 2d since early summer, i'd also suggest checking out animics and ponyfaetality.
Somehow v2 works better for me. I love this! :)
Hi, people! How to remove constant transformation into insect when using tag "insect".
I need a person with a bee for example, NOT a bee-person. Negatives - source_pony or source_furry - do not give results.
are you trying to create a person holding or banging a bee?.
@skullx198x980 And here it doesn't matter. "Standing nearby, flying nearby, just somewhere in the background, etc., etc." But with any mention of any type of insect (be it a bee, wasp, ant, etc.) - it turns the character into it, and does not add it as a separate one.
It almost sounds like a bug
@dimirbovg i'd maybe check your negs or the words in your prompt, start off with things only related words.
If you messed with token merging that might mess with it not sure.
pretty much got what you were looking for right away every pic had a bee flying around with various sizes and samus standing
used cfg 5, 30 steps,, 896x11523, dpm++2m sde karras
pos score_9, score_8_up, samus standing, small bee, bee on hand,
neg tail, stinger,
prefectponyv3
@skullx198x980 In general, there were two problems:
1. I had a different sampler.
2. The problem was in the LORA, there were bugs
Thanks for the answer, it helped in principle.
@dimirbovg no problem glad to hear you were able to resolve it
great stuff!
It's the best model, thank you!
This is my favorite model, it's just...chef's kiss.
It's super good <3 I love it.
Overall style and character face come out super nice n super sharp. It's better than AutismMix.
But there are few issues...
When I prompt "pregnant" the image sometimes come out with a pregnant female slightly lactating or pussy juice coming out from her vagina even without the existence of this tags within the whole prompt. So I guess there are bit of contamination.
one of my favs
Surprisingly a versatile model that could generate a couple of different anime artstyle! 11/10
nice model，I used it to create an image that I'm quite satisfied with.
like
This model surpassed Autismmix a long time ago. You should be right there next to PDXL. Give the man more thumbs up!
"added some style to it"
Meaning some styles it can recreate or a style that goes with it by default in every image it generates?
inbuilt to enhance the base pony style and whole image
Any chance of a clean variant of v3 in the future?
working on v4 and v4 lighting
This is my favorite model, I always use it
How do I use with Draw Things app?
Details
Files
prefectPonyXL_v3.safetensors
Mirrors
prefectPonyXL_v3.safetensors
prefectPonyV3_v10.safetensors
prefectPonyXL_v3.safetensors
rrbTest01_rrbTestV10.safetensors
prefectPonyXL_v3.safetensors
prefectPonyXL_v3.safetensors
prefectV3_v3.safetensors
prefectPonyXL_v3.safetensors
prefectPonyXL_v3.safetensors
prefectV3_v3.safetensors
prefectPonyXL_v3.safetensors
prefectPonyXL_v3.safetensors
prefectPonyXL_v3.safetensors
prefectPonyXL_v3.safetensors
prefectPonyXL_v3.safetensors
prefectPonyXL_v3.safetensors
pref.safetensors
prefectPonyXL_v3.safetensors
prefectPonyXL_v3.safetensors
prefectPonyXL_v3.safetensors
prefectPonyXL_v3.safetensors
Available On (3 platforms)
Same model published on other platforms. May have additional downloads or version variants.