!!! UPLOADING/SHARING MY MODELS OUTSIDE CIVITAI IS STRICLY PROHIBITED* !!!
Check my EXCLUSIVE models on Mage.Space: AniMage PXL • AniReal PXL • Lucid Dream • AniMage SD1.5 • Realistic Portrait
SDXL - Pony: AniVerse PXL • AniMerge PXL • AniToon PXL • AniMics PXL • AniVerse XL
SD1.5: AniVerse • AniThing • AniMerge • AniMesh • AniToon • AniMics
Also in Collaboration with Shakker.ai
This model is free for personal use and free for personal merging(*).
For commercial use, please be sure to contact me (Ko-fi) or by email: samuele[dot]bonzio[at]gmail[dot]com
⬇Read the info below to get the high quality images (click on show more)⬇
Aniverse - Pony XL - make the impossible possible!
This is a long shot project, I’d like to implement something new at every update!
The name is a merge of the two words: Animation and Universe (and a word pun: Any+Universe -> Anyverse -> Aniverse)
-> If you are satisfied using my model, press on ❤️ to follow the progress and consider leaving me ⭐⭐⭐⭐⭐ on model review, it's really important to me!
Thank you in advance 🙇
And remember to publish your creations using this model! I’d really love to see what your imagination can do!
Recommended Settings:
Excessive negative prompt can makes your creations worse, so follow my suggestions below!
Before applying a LoRA to produce your favorite character, try it without first. You might be surprised what this model can do!
Best CONFIGURATION that I found
VAE: a special VAE is already included, you don't need it - (Thanx to nuaion)
Upscaler: 4x-Ultrasharp or 4X NMKD Superscale
Clip skip: 2
Width: 768
Height: 1344
CFG: 5.5
Steps: 30
Sampling favorite method: DPM++ 2M or Euler Max
Scheduler: Karras
EXAMPLE OF GENERAL PROMPT:
POSITIVE PROMPT: score_9, score_8_up, score_7_up, (Type of Shoot), (Subject), the description, (background), 4n1v3rs3, more details
EXAMPLE: score_9, score_8_up, score_7_up, Portrait, yorha no. 2 type b, short white hair, black dress, hairband, clothing cutout, cleavage cutout, puffy sleeves, black hairband, feather-trimmed sleeves, juliet sleeves, mole under mouth, sunset, rubble in the background, 4n1v3rs3, depth of field, dynamic angle, fashion photography, sharp, hyperdetailed:1.15
NEGATIVE PROMPT: score_6, score_5, score_4, worst quality:1.4, low quality:1.4, front light, grayscale, doll, plastic, fake, ugly, hair on face, muscolar woman, low res, blurry, fat, topless, child
Stable Diffusion XL with only 3GB of VRam (nVidia GPU):
If you have a GPU nVidia, but do you have problem to run XL cause low VRam, try to use this version made by me and nuaion.
It is a Portable version, so it does not affect if you have an A1111 already installed on your PC and can easily work in parallel.Besides the fact that it is "movable" from PC to PC without problems.
This version does not have any pre-installed template nor ADetailer which we preferred to leave optional
Let me know if it works for you and what you thinkA1111 my settings:
I run my Home PC A1111 with this setting:
set COMMANDLINE_ARGS= --xformers --skip-torch-cuda-test --no-half-vae
(if you have low VRam, try to add --medvram-sdxl or --lowvram that can help you, but it slow down the image creations)
if you can't install xFormers (read below) use my Google Colab Setting:
set COMMANDLINE_ARGS= --disable-model-loading-ram-optimization --opt-sdp-no-mem-attention --no-half-vae
(if you have low VRam, try to add --medvram-sdxl or --lowvram that can help you, but it slow down the image creations)
My A1111 Version: version: v1.9.3 • python: 3.10.11 • torch: 2.1.2+cu121 • xformers: 0.0.23.post1 • gradio: 3.41.2 •
If you want activate xformers optimization like my Home PC (How to install xFormers):
In A1111 click in "Setting Tab"
In the left coloumn, click in "Optimization"
in: "Cross attention optimization" select: "xformers"
Press in "Apply Settings"
Reboot your Stable Diffusion
If you can't install xFormers use SDP-ATTENTION, like my Google Colab:
In A1111 click in "Setting Tab"
In the left coloumn, click in "Optimization"
in: "Cross attention optimization" select: "sdp-no-mem - scaled dot product without memory efficient attention"
Press in "Apply Settings"
Reboot your Stable Diffusion
How to emulate the nvidia GPU follow this steps:
In A1111 click in "Setting Tab"
In the left coloumn, click in "Show all pages"
Search "Random number generator source"
Select the voice: "NV"
Press in "Apply Settings"
Reboot your Stable Diffusion
If you use my models, install the ADetailer extension for your A1111.
Navigate to the "Extensions" tab within Stable Diffusion.
Go to the "Install from URL" subsection.
Under "URL for extension's git repository" put this link: : https://github.com/Bing-su/adetailer
Click on the "Install" button to install the extension
Reboot your Stable Diffusion
How to install Euler Max Sampler:In A1111 click in "Extensions Tab"
click in "Install from URL"
Under "URL for extension's git repository" put this link: https://github.com/licyk/advanced_euler_sampler_extension
Once installed click in: "Installed" Tab
Click in "Apply and quit"
Reboot your Stable Diffusion
Now at the end of the list of the sampler, you have the new sampler.
HiRes.Fix Setting:
I don't use Hi.Res fix because:
1) in my computer don't work
2) my models don't need it. Use txt2image, aderailer and the suggested upscaler in the resources tab.
If you still want use it, this is the setting sent me by MarkWar (follow him to see his creations ❤️).
Hires upscale: 1.5
Hires steps: 20~30
Hires upscaler: R-ESRGAN 4x + Anime6B,
Denoising strength: 0.4
Adetailer: face_yolov8n
How to install and use adetailer: Click Here
Here you have a review (in spanish) of AniVerse Pony XL Model (thanx to Salió Aniverse XL | Stable Diffusion en español)
Do you like my work?
If you want you can help me to buy a new PC for Stable Diffusion!
❤️ You can buy me a (Espresso... I'm italian) coffee or a beer ❤️This is the list of hardware if you are courius: Amazon Wishlist
I must thank you nuaion and GattaPlayer for their support
You are solely responsible for any legal liability resulting from unethical use of this model
(**) Why did I set such stringent rules? Because I'm tired of seeing sites like Pixai (and many others) that get rich on the backs of the model creators without giving anything in return.
(***) Low Rank Adaptation models (LoRAs) and Checkpoints created by me.
As per Creative ML OpenRAIL-M license section III, derivative content(i.e. LoRA, Checkpoints, mixes and other derivative content) is free to modify license for further distribution. In that case such is provided by licensing on each single model on Civitai.com. All models produced by me are prohibiting hosting, reposting, reuploading or otherwise utilisation of my models on other sites that provide generation service without a my explicit authorization.
(****)According to Italian law (I'm Italian):
The law on copyright (law 22 April 1941, n. 633, and subsequent amendments, most recently that provided for by the legislative decree of 16 October 2017 n.148) provides for the protection of "intellectual works of a creative nature", which belong to literature, music, figurative arts, architecture, theater and cinema, whatever their mode or form of expression.
Subsequent changes, linked to the evolution of new information technologies, have extended the scope of protection to photographic works, computer programs, databases and industrial design creations.
Copyright is acquired automatically when a work is defined as an intellectual creation.
Also valid for the US: https:// ufficiobrevetti.it/copyright/copyright-usa/
All my Stable Diffusion models in Civitai (as per my approval) are covered by copyright.
Description
My personal page for donation (and where you can buy my model in early preview): Samael1976 Ko-fi page
What's new:
As I write this message (which for me is early January) I am doing the longest training I have ever done for a model (
it started on December 26th and will last for 45 consecutive days). So, not knowing if I would have had the time to train a new version of AniVerse, I preferred to cover my back, and create this new version that is effectively a merge between version 3.0 and version 4.0. In this version I tried to take the best things from both models and implement it in this one.I don't know if the new model will come out right after this version of AniVerse, above all I hope I haven't thrown away 45 days of training!!!
EDIT: Unfortunately, I only realized after over 20 days of training that I had made a mistake in the settings.
So I have to start everything again. The positive side is that I'm adding more things, the negative side is that the training time will get even longer
BEST CONFIG:
VAE: a special VAE is already included, you don't need it - (Thanx to nuaion)
Clip skip: 2
Width: 768
Height: 1344
CFG: 5.5
Steps: 30
Sampling favorite method: DPM++ 2M or Euler Max
Scheduler: Karras
EXAMPLE OF GENERAL PROMPT:
POSITIVE PROMPT: score_9, score_8_up, score_7_up, (Type of Shoot), (Subject), the description, (background), more details
EXAMPLE: score_9, score_8_up, score_7_up, Portrait, yorha no. 2 type b, short white hair, black dress, hairband, clothing cutout, cleavage cutout, puffy sleeves, black hairband, feather-trimmed sleeves, juliet sleeves, mole under mouth, sunset, rubble in the background, 4n1v3rs3, depth of field, dynamic angle, fashion photography, sharp, hyperdetailed:1.15
NEGATIVE PROMPT: score_6, score_5, score_4, worst quality:1.4, low quality:1.4, front light, grayscale, doll, plastic, fake, ugly, hair on face, muscolar woman, low res, blurry, fat, topless, child
Very Important Thing:
I finally found out what causes aDetailer errors: it's parentheses()
So if you want to avoid face or image generation errors, avoid using (or use them as little as possible) on both the positive and negative prompts
FAQ
Comments (62)
It's done amazingly for me.
thank you so much!
Is version 4 going available for on-site use?
I just requested to move from version 3.0 to 4.0. TheAlly, (the person from Civitai who usually takes care of it) is always very fast and helpful. However, consider that it is Sunday and it is a holiday for him too ;)
@Samael1976 thanks I can wait let him have his holiday
@Blindparty done!
@Samael1976 it keeps focusing on portraits even with full body shot in the prompts
@Blindparty That's strange, I haven't had this kind of problem in 2400 generations... so I'm inclined to think that there's something wrong with your prompt... try using my image metadata.
@Samael1976 I thought so too so i tried follow you're layout with this and still came out with a portrait.
score_9, score_8_up, score_7_up, full body shot, Changeling, pale skin, white eyes, white hair, dark eyelids, black eyeshadow, black lips, black lipstick, side part hair, short hair, hair just above shoulders, medium breasts, testicles, attractive face, cute face, curvy body, big hips, wide pelvis, slim waist, medium penis, nude, black bed, black decorations, black bedroom background, 4n1v3rs3, depth of field, dynamic angle, fashion photography, sharp, hyperdetailed:1.15, high-resolution, Expressiveh, masterpiece, best quality, dgirl, shemale, newhalf, transgirl, dickgirl, girl, uncensored, solo, laying on back, moaning, masturbating,
@Blindparty in the prompt there are a lot of contradictions, which are definitely not liked and double repetitions. Example, on one side you say short hair, on the other you say hair above the shoulders. Either one or the other. Besides the fact that everything regarding "masterpiece, high quality" pony (the generic model) does not know how to interpret it. Anyway, try putting: barefoot after your nude.
besides I have serious doubts that this prompt exceeds the 75 tokens which is the structural limit of the XL/Pony models
I can't test because I'm doing a training that will end in 33 days...
@Samael1976 I never knew about the 75 token limit. cuz no other checkpoint has done what yours has ever done completely ignore part of the prompt to the point of where I had to ask what was going wrong. Even if I can't get yours to work for me it has helped me learn how to structure the prompts a little better. So you have my thanks
@Blindparty Hello my friend, first of all... Happy New Year!!! There would be a lot to write... know that the base of the model (in this case XL, because Pony is a fork of XL) has its own limits. For XL, there are 75 tokens, I honestly don't remember how many there were for the SD1.5 version. The Pony prompt differs from XL for the wording of the scores. If with classic XL you can use masterpiece, high quality, for Pony they can be used, but for him they have no meaning, because he was trained using another system. For this reason, score_9, score_8 etc. etc. are used, because they are equivalent to choosing only "the best base" with which pony was trained. I can anticipate that I am training a basic Pony model right now (it will take me a while... 45 days of computer working 24 hours a day). But I'm training it so that it should be able to exclude scores and be able to insert masterpiedce, high quality... the model, if it will work, will be: AniThing because it will use 4 different concepts and you will be able to recall the various styles (realistic, semi-realistic, anime and comics) simply by using a specific trigger word.
@Samael1976 I'm surprised no one puts that on their checkpoints that there's a token limit. Because I learn by following the prompts that people made on site and that's essentially a paragraph
@Blindparty You have no idea how many errors in the prompts I have seen. Most people have not "studied". I still make mistakes, after spending months studying the subject. However if you search the internet for "stable diffusion tolen limits" you will find a lot of documents about it.
@Blindparty I'll give you a small example. Many people only use Euler A, or don't follow what the author (trainer) writes as recommendations for their model. Which is a huge mistake. Because each trainer customizes the training based on his tastes and what he wants to achieve. There is no "standard" valid for all models. But you should always follow what the author writes in the notes, to get the best results
@Samael1976 yeah I've been reading the pages that is why i was surprised when you said 75 tokens cuz no one ever mentions it. I guess they expect you just to know it. but I didn't know changing Euler A could affect the results
SIIIIIII lo stavo aspettando più del Natale xD TE SE AMA
lol, thanks! I was really hesitant to change the schedule and put it online on the 35th as a Christmas present... but it would have been quite complex, changing the date of all the uploaded galleries.
@Samael1976 Cade a fagiolo perché stavo preparando il dataset per fare un altro LoRa (di Sette di Nove di STVOY), stavo smattando per correggere gli occhi perché su Pony come PG lei non esiste, adesso rigenero tutto con la v4, però a sto giro mi faccio lo stylesheet e non me ne esco, quello di Lara funge per miracolo ._.
@huchukato ok, ma la prossima voglio T'Pol!!!
@Samael1976 E' in lista anche lei ma mi devo fare tutto il dataset da zero, facendo gli screen dalle puntate (na scusa buona per riguardare ENT). Comunque sta cosa che in Pony non esiste ST è MALE.
Fast favorite. Thank you! This model has a great balance between light and shadow, clarity without being too realistic and reasonable prompt adherence.
Thank you❤️🤗
Often blue clothing elements predominate. Explain the difference between AniMix and AniToon?
Just use more words for description of clothes. Do you really don't see the difference from AniToon and AniMics? AniToon is anime japanese style, AniMics is a western style comics/animation. Just read the full models descriptions, is all written
@Samael1976 Thanks for the clarification, I was looking for a difference in drawing quality.
Well I find V3 better for my intended purpose, but still great.
I don't know your intentions, but anyway, thanks 🤗
@Samael1976 I use CLIP SKIP 1 for more hybrid Anime style which V3 seems to handle a bit better.
Though I noticed CLIP SKIP 2 also messes up straight lines behind characters in V3 and V4, so I still default to 1.8 if I want less 2d look.
PS: Also Lower CLIP SKIP tends to give you more detailed background albeit at the cost of introducing elements you don't want.
@aerd222 Thank you! Your feedback will help me a lot, thank you so much!
@Samael1976 No problem, overall I would sum up the difference between CLIP SKIP 1 and 2 is that:
the Higher CLIP SKIP tends to give you smoother image with less detail.
The lower CLIP SKIP tends to be more detailed at the cost of requiring you to fiddle more with the prompt to get what you want.
Also, I found that it is better to use HI Res with 15 steps and 0.5 strength. At least in my case. More steps don't seem to improve quality after 15. You could also use hi-res with DPM++ 3M for slightly different results.
And if you need to be quick you can go to as low as 5 steps, but then it's better to use SD Latent X2 for upscaling. And only DPM++ 2M for hi res. You also might want to increase strength to 0.55 .
But generally 15 steps is optimal, 10 is decent and 5 is mostly when you need to churn something really fast.
I would upload some images, but apparently when I upload them they get recognized as duplicates and are deleted.
How do you change the character's face with this model? The rendering is very nice but the face is always pretty much the same one (child-like Asian face), no matter which Lora I select. Any ideas?
if you look at the user submitted images, you will find a lot of examples of Lora applied and working. So, either you have some problem using the software for generating images, or you have the seed blocked, or the lora you are using, are not good lora. Take a look and let me know! Cheers!
I made a post showing the effects of a LoRA on the base model ( https://civitai.com/posts/10956955 ).
There are some things you can do with prompting as well, but AniVerse has a pretty strong style which is popular.
As for LoRAs.They have been generally working for me at strengths 0.6 - 0.7.
In your specific case, without getting a look at your prompt or workflow, there's only so much advice one can give.
@pope_phred Tha k you Pope for your help! I really appreciate it! 🤗
when will it be available for generation on civitai? :-)
Thanks to @theally it's already done! 🤗
Thank you for all your great work. Happy New Year!
Thank you so much! I Really Appreciate! And happy new year you too! 🤗
No me canso de usar este checkpoint, para mi no es que sea el mejor no, ni lo siguiente no, ni lo plus ultra, es que se ha convertido en uno de mis dos checkpoints fijos, úsalo, es perfecto para todo!!! Gracias Samael!!!
¡Gracias amigo mío, dedicaste unas palabras maravillosas y te lo agradezco desde el fondo de mi corazón! ¡Te deseo un maravilloso año próximo! 🤗🤗🤗
hey should i use the trigger word "4n1v3rs3" when generating with this model???
Yes, it's better because it take more the style of AniVerse
@Samael1976 oh man i use it wrong then already hundreds of images generated without the trigger word 😭
Gonna test with the new version then, thx for the reply.
@fronyax it's not a big problem really! You are welcome
PS: little spoiler I'm training now (45 days of training) a new model (not aniverse) with 4 different concept style. In this future model if you don't use the trigger word for recall the style, it mix the 4 style in 1). So in the future model, use the trigger words, it's a mandatory ;)
@Samael1976 Yo have you ever had a problem where suddenly every pony model you use starts producing very noisy, garbled up images that are partially burned and deep fried?
This started for me when i aborted a generation one too many times or something a couple days ago.
And now, every prompt/lora/weight combo that used to produce quality results, produces very low quality stuff filled with artifacts.
I ended up full uninstalling Python and Git and SD, then i switched to Forge from A1111, and still the same problem persists across every model i try to use.
I searched probably the entire internet and no setting or prompt change fixes this, except for reducing the number of LORAs involved down to 3 max. Only in this case is the generation free of noise and artifacts.
But it makes no sense because I used to deploy up to 20 LORAs with any Pony model and it would still produce quality.
Wondering if you would know anything about this sort of phenomenon?
Anyway good luck with your work tho
@GardenHug I'm so sorry, but I haven't heard or seen anything about this phenomenon. Besides the fact that as you can see from my samples, I never use loras... If it can console you, some time ago I wasted a week trying to install and uninstall various versions of A1111, but nothing worked and I had to format the entire PC. The only thing I can recommend is to try my portable version of A1111. It's a completely standalone version and already has everything you need to run it completely independently from other versions of A1111/Python/git. You can find it by looking at my articles. The one that says how to run XL with only 3gb of vram. Obviously, when you start it, use the option without vram limitations. Please, let me know, I'm curious!
@Samael1976 ight so i still don't really know wtf had happened, but the main culprits seem to have been some SDXL loras.
Some were glitching outright so they were removed, others are still fine, and one in particular (DPO (Direct Preference Optimization) by 'enfugue' now needs to be set at no higher than 1.00 weight.
Previously, i had generated just fine with this LORA being set at 9.00.
The other LORAs that now seem to be broken have also never caused problems before this.
Redownloading them didn't help.
I seem to have solved the problem for now.
I also ended up switching to ReForge, it's way faster
my favorite model❤️❤️
Thank you!!!
Great aesthetic. Good detail without clutter. Nice work!
i love your models dude, but V3 still rocks hard !!
I agree. One thing I noticed is that 4.0 and 5.0 are not so versatile when coupled with enhancement Loras by Velvet and Vixon. For example, I'll prompt for a lady doing a standing split, or hands behind head and I keep getting a close up shot. But with 3.0... it works great. Those Loras work well with other pony checkpoints I've tested, but I like the model in Aniverse the best.
Thank you for sharing your solution for the aDetailer problems! Huge help!
Hello Samael, do you plan to release a version of this checkpoint but in Illustrius?
at the moment, no. Because I'm training a new model... and my PC is running 24 hours per day... and the training least more or less 3 months
Thank you so much for this! more success to you!
thank you! ❤️
Beautiful work. Thank you.
Thank you so much ❤️
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.