You can also run this model on Magespace and Sinkin.ai! https://www.mage.space/
V5 - https://www.mage.space/play/9200bfc93333da71123999f3550aabaa
V7 - https://www.mage.space/play/0a2a61c6d4a668ef51f552df0231067d
https://sinkin.ai/m/RR7Vrj4
Want to send some support? (Buy a cup at Ko-fi)
FYI VAE IS BAKED IN
Version 3 has https://huggingface.co/madebyollin/sdxl-vae-fp16-fix baked in.
RealCartoon-XL is an attempt to get some nice images from the newer SDXL. It still is a bit soft on some of the images, but I enjoy mixing and trying to get the checkpoint to do well on anything asked of it. This model level is definitely pushing my computer, so takes a bit longer to actually get it things tested and mixed :)
The mission is the same of the SD1.5 versions. Which is to get a checkpoint to pretty much do well on prompts while looking good with good variety
I hope you all enjoy it! Please review and share your images. I very much appreciate the support with the downloads and feedback (THANK YOU ALL). Never thought it would get this much attention.
The Process:
The starting checkpoints for merging were a couple of top ones of course (The checkpoints do/did not have restrictions on checkpoint mergers. Some of these are in my recommended resources as well). I also baked in the VAE (sdxl_vae.safetensors).
Currently this checkpoint is at its beginnings, so it may take a bit of time before it starts to really shine. I will continue to update it as time progresses...but I do hope you all enjoy it as it develops.
The Settings:
Still figuring out SDXL, but here is what I have been using:
Width: 1216 (normally would not adjust unless I flipped the height and width)
Height: 832
Sampling Method: "Eular A" and "DPM++ 2M Karras" are favorites.
Sampling steps: 30 - 55 normally (30 being my starting point, but going up to 55 and even 60 a lot of the time)
Hires.fix settings: Upscaler (R-ESRGAN 4x+, 4k-UltraSharp most of the time), Hires Steps (10), Denoising Str (0.34 - 0.45 normally), Upscale (1.5 or 2 does well)
Clip Skip: 2
Some settings I run on the web-Ui to help get the images without crashing:
set COMMANDLINE_ARGS=--medvram --no-half-vae --opt-sdp-attention
__________________________________________________________________________________________________
License & Use
This model is open access and available to all, with a CreativeML OpenRAIL-M license further specifying rights and usage.
1. You can't use the model to deliberately produce nor share illegal or harmful outputs or content.
2. The authors claims no rights on the outputs you generate, you are free to use them and are accountable for their use which must not go against the provisions set in the license.
3. You may re-distribute the weights. If you do, please be aware you have to include the same use restrictions as the ones in the license and share a copy of the modified CreativeML OpenRAIL-M to all your users (please read the license entirely and carefully).
Please read the full license here Stable Diffusion
Use Restrictions:
You agree not to use the Model or Derivatives of the Model:
- In any way that violates any applicable national, federal, state, local or international law or regulation
- For the purpose of exploiting, harming or attempting to exploit or harm minors in any way
- To generate or disseminate verifiably false information and/or content with the purpose of harming others
- To generate or disseminate personal identifiable information that can be used to harm an individual
- To defame, disparage or otherwise harass others
- For fully automated decision making that adversely impacts an individual’s legal rights or otherwise creates or modifies a binding, enforceable obligation
- For any use intended to or which has the effect of discriminating against or harming individuals or groups based on online or offline social behavior or known or predicted personal or personality characteristics
- To exploit any of the vulnerabilities of a specific group of persons based on their age, social, physical or mental characteristics, in order to materially distort the behavior of a person pertaining to that group in a manner that causes or is likely to cause that person or another person physical or psychological harm
- For any use intended to or which has the effect of discriminating against individuals or groups based on legally protected characteristics or categories
- To provide medical advice and medical results interpretation
- To generate or disseminate information for the purpose to be used for administration of justice, law enforcement, immigration or asylum processes, such as predicting an individual will commit fraud/crime commitment (e.g. by text profiling, drawing causal relationships between assertions made in documents, indiscriminate and arbitrarily-targeted use).
Terms of use:
- You are solely responsible for any legal liability resulting from unethical use of this model(s)
- If you use any of these models for merging, please state what steps you took to do so and clearly indicate where modifications have been made.
Note:
If you see any conflicts or corrections to be made, please let me know.
Description
This one was interesting to work on. My computer did not want to work with me lol; but finally started working. So, for this version I am just trying to improve the overall look and response. SDXL is rather cool in that it does better with it range based on the prompt...but I still like the creative look SD 1.5 does initially (well at least with my other RealCartoon checkpoints that is, lol). ANYWAY, this version responds well and does well with details....and if you prompt for more of a realistic look it does well. Anyway, I hope you all enjoy!! Last update for the year :P ....my be busy with new child at this point...so may not get to update for a minute.
FAQ
Comments (78)
Was trying to wget this to my Runpod. Since Civil.ai doesn't have password anymore I can't, wget with credentials, and cookies doesn't work either... I tried uploading it manually but it's just too much.
Did you try the 6G version? Also I know it requires login normally to download…so unsure if that causing issue.
@7whitefire7 It was causing the issue. I can usually enter my credential in wget to bypass that, but I never had a password for Civic.ai, just a google account. When you do reset password, they just send you a mail to login ... I uploaded the model to a GDrive, and manage to download with a special command.
Whenever you have problems like this (regardless of the platform - civit or otherwise), the normal trick is to start the download on your computer, then go into your browser downloads listing, and right click the URL that is being downloaded, and copy it to clipboard, and then run `wget -O foo.safetensors "the url from your clipboard"` (make sure you includes the quotes around the URL) on your runpod machine, or wherever. The URL will have the required credentials embedded, which is why it works. This also works with Google Drive downloads, for example - and basically every other platform that tries to prevent you from getting a direct download URL.
Why are there 2 models to download? One is 6GB, the other 12GB.
Large one is for merges :). Both generate the same images though.
@7whitefire7 What is a merge? :)
@Pete14123 Taking multiple checkpoints to create a new one.
I'm liking the flexibility in the model BUT i need some advice when using a trained Lora with this model. I seem to lose recognisability in my Lora if I use strong weights...should i retrain my Lora using this as the base?
I am using 1024x1344 for portraits and do not get enough space between the top of the head and the top border. I there a prompt to prevent this?
I did some onsite generation testing and putting zoomed out portrait should leave more room. I will have to test more prompts to see response, but I am away from computer…but hopefully that helps you.
I don't think "zoomed out portrait" makes a difference. Maybe slightly better.
@Pete14123 will see what I can do. Thank for letting me know.
When I try to Download this model there are 2 versions, one is around 6 GB while the other is around 12. Could someone let me know the difference and whether it is worth it to download the larger version? Thanks!
Smaller one is fine for image generation. Bigger one is more useful for merges :)
@7whitefire7 Thanks!
Great model, keep up the great work!
All of them great. like out of the park 10 out of 10
I know its new and just a question are you thinking of doing a Turbo Model. Just a question. p.s Thanks
Thank you. I hope to get into the Turbo area. Need to actually sit down and focus on it lol. :)
nice
Why are there two models 12gb and 6gb, so what is their difference?
One is better for merging and one is just pruned down. They both pretty much do the same on image production.
@7whitefire7 If I don’t do training, is there still any reason to pick the full checkpoint? Will it lead to minor quality improvements?
@Jaidchen Not really. I did not notice a difference. I normally use pruned version due to lack of space on my drive lol.
Waited overnight for this to properly load in my Stable Diffusion, 12 hours later it still has that orange spinning thing and only renders abstract art. Am I doing something wrong?
? You using A1111?
@7whitefire7 Yes. the little orange thing that says processing doesn't stop.. Extensions never worked for me either, not sure if that's connected to the problem. How long is it usually supposed to process for?
@sonicasino237 Have you used other SDXL or is it just this one that is causing issues (if that is the case i would just redownload as something may have happened in the download). I use A1111, and it does not take long to load. I have it updated though....so usure....I am using V1.6.0 with python V3.10.6.
@7whitefire7 I think I ha to downgrad python to get other things working, but yes some other ones are working. I did try downloading it twice. The last 2 times, the screen turned blue and it said there was a serious error and it has to restart the machine o_o
@sonicasino237 :( that is weird. I hope that fixes it. Does the A1111 show any errors at all?
@7whitefire7 It did say error at one point, not sure what it was referring to. I'll give it a few more tries, or try again once I get a new computer. Or maybe it's a A1111 problem? Are there easier ones out there? The node one looks confusing.
@sonicasino237 Well A1111 is the easier one that I have seen. Comfy UI has more options, but can be confusing. Maybe put “set COMMANDLINE_ARGS=--medvram --no-half-vae --opt-sdp-attention” without the quotes in the startup bat file. May help.
@7whitefire7 Thanks I'll try
Why are there two version? 6gb and 12gb?
Larger good for merging and images. Smaller does same on images but saves space.
What is new in v6?
It has better prompt response, better overall image quality (in my opinion) and more versatility :).
This is an excellent checkpoint for all kinds of 2d art and illustration. It’s able to respond not just to descriptions of art styles (ie 2d manga), but styles of artists (ie Rembrandt), and even other works (ie Indiana jones). I basically don’t use any other XL checkpoint at this time. The only thing I would love to see added is better support for overt nsfw actions- right now the model is excellent at standing around and posing, but not very versatile with more complex actions. However, it can’t be denied that the aesthetic qualities of this model are stunning.
Ty for feedback.
The full model is provided for merging experiences. Merging and testing takes a lot of time and consumes power. If you don't like what you download and want more of "this" and less of "that": merge your own models.
@hansolocambo I don’t know how you could read my comment and see it as anything but a positive review 🤔
I looked through the documentation and comments. The results tend to come out looking realistic and without any cartoon influence. Is there a specific keyword to help influence the output?
Using words in the prompt like: Anime, Animated, Cartoon, Comic, painted, or something along those lines normally works. I am trying to get a mix in there....the goal is a balance between the two...but SDXL can handle a larger range then SD 1.5
@7whitefire7 - Thanks. I love your SD content and the XL model is fantastic as well. I appreciate you! I seem to either get realistic images or compete cartoon images. The balance is certainly hard to nail!
It's easy to stumble upon quality images by just typing "ldrkjglk", "klgjkrdl", or even "ldrkjglk klgjkrdl". I haven't tested other combinations yet... Well, like entering normal human words.
Is "lol" a normal human word?
yea I have done one word just to see what it does lol....can be fun :)
@VirtualSun Well, that depends on what is defined as "normal". It is understood today, and thus might have been put in a computer code....sure. Is it something that is known to much of human history....no....so not normal then. :)
Any chance to get the smaller version for v6 as well?
12gb for a model is a bit too much T.T
I show both files as downloadable? Do you not have the option to download the pruned version?
@7whitefire7 ah, you're right!
on the previous version it shows the two option on top as well, why on version 6 you have to scroll down the file section to see the smaller version. But it's there, my bad!
@naughtyskynet No worries, that is weird it not showing up on a pull down.
IMO, the default results from this are a bit TOO real with not enough cartoon, usually. They often just kind of look like real photos with an "airbrush" sort of filter. I think the overall style of RealCartoon 3D V14 is a bit better.
Thank you for feedback :)
@7whitefire7 any plans on making a realcartoon3d for SDXL?
@pohttems Hoping to. Just getting the time and computer power is my problem lol. I hope to though.
@7whitefire7 what would you need in terms of hardware for making it? I can imagine the VRAM requirements are higher than SD 1.5
If you don't have DPO, start using it in your XL models, I'll leave you a list (a copy and paste of my comment I made in another post) maybe it can help you to improve your XL models.
1-AutismMix SDXL
https://civitai.com/models/288584?modelVersionId=324619 (all three of their model mixes have DPO)
2-Confetti Comrade Mix
https://civitai.com/models/280344/confetti-comrade-mix(here I had left some comments giving ideas to move the community forward, anyway I will leave my ideas in this post in case someone reads it and wants to know more about it)
3-Js2Prony
https://civitai.com/models/280321/js2prony
4-Comradeship XL
https://civitai.com/models/246299/comradeship-xl(my favorite so far)
5-[0.8(ponyDiffusionV6XL_v6) + 0.2(reproginexl_v10)] > NAI 3 !
https://civitai.com/models/263797/08ponydiffusionv6xlv6-02reproginexlv10-greater-nai-3
6-Hassaku XL
https://civitai.com/models/140272/hassaku-xl-sfwandnsfw-beta(this model understands prompts well but is not good at replicating characters without loras,the other models in the list including the Pony Diffusion V6 XL can replicate characters perfectly,even better than Loras already made in the civitai community)
the list I just left have good implementation of (DPO) the idea of this post is that the creators already famous in civitai begin to implement this technology,98% of civitai models are mediocre but with this technology the existing models can make magic,genitals,legs,hands,expressions,poses, concepts etc make more sense with these models already named(Pony Diffusion V6 XL was the first to implement the technology to a XL model) but the ones already named here are of very good quality also it would be great to continue seeing models with many different styles and more polished in the future) researching I discovered that if you mix a model that already has(DPO) with another model already existing civitai gives results like these. ..the technology works very well with XL but as far as I can see very bad with 1.5(for the moment) I think the future this year is XL I think the other model and lora creators should all work for XL,it feels good when you put indications and you get almost everything as you want,that with 1.5 is impossible,I am using XL,without controlnet and without Lora I have had incredible results,the mixes with the DPO models I think it is better to do them in Comfyui,I hope this post can help you,there are many people who still don't know about the technology
I took the time to test about 50 XL models and all were exactly the same as the 1.5,bad anatomy etc,in search of DPO models I did my tests by replicating famous characters,an effective way to test is using as a prompt:
cammy,chun li,jinx,league of legends,etc,if characters like that replicate perfectly the model has DPO,if the replica of the character doesn't come out then it doesn't,the models in this list can replicate the characters already named and many many more,(just do your list of character tests)....
PS:there are models in civitai that claim to have DPO and don't have them or are very bad,all of them were tested by me so I save you time
here you can read the post of this page to know a little more about DPO
https://civitai.com/models/242825/dpo-direct-preference-optimization-lora-for-xl-and-15-openrail
Ok. I admit that I didn't know the (--opt-sdp-attention) command line. And this has literally changed the way I work, because now I can invest most of my time in the XL models, without my potato computer exploding.
:) can be very helpful depending on the computer.
Big fan. In addition to pretty accurate prompt interpretation, it's also the base ckpt that gives me the best chance for realistic skin.
:) glad you are enjoying!!
I'm in love with model! It follows the prompt so closely!
Please make more SDXL content!
Thank you for the pruned version. My device do not support unpruned version.
Welcome :)
One of my favourite models!
It fails to load or convert. What's the reason?
I would try a re-download if having issues loading it. If on site, then that a Civitai issue :/
What pixel resolution does this model like?
From the description:
Width: 1024 (normally would not adjust unless I flipped the height and width)
Height: 1344 (have not done too much higher at the moment)
I noticed this checkpoint knows much celebrities, but you will get a real somigliance only repeating the name also in the ADetailer prompt.
For example, if prompt is: "1girl is Linda Carter sitting on armchair, futuristic dress, etc...
You have to write "Linda Carter" also in the Adetailer box.
But, at this point there is a big problem: what happen is you need 2 or more people in the image? They will have the Linda Carted face, all of them!
So, someone knows a trick to say Adetailer to apply a differentiate solution?
regional prompter
@jjooosss755 Unfortunately after using "regional prompter" since years, (often getting more delusion than benefits...) I noticed it considerably lower the entire image quality. With SD 1.5 this fact was no so noticeable (may be because 1.5 is unprecise by nature) But on XD/Pony, you see a clear difference, so I dont use it more...
If Adetailer cant differentiate, the "inpaint" is the real solution...
Thanks for creating RealCartoon-XL checkpoints and making it available to us. It does wonders. Awesome image creation. Here's my tip.
Thank you, and glad you enjoy the checkpoints :)
Terrific model !! So versatile… and getting better all the time ! Thank you for making this available and perfecting it over and over again ! ;-)
this model is so cool. please don't mind me spamming images here... happens!?! :3
I do not mind....I love seeing what people generate with it.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.