Follow me on X/Twitter for exclusive content: https://x.com/Citron_Legacy 馃榿
Wield a rainbow's radiance, bright and flat,
Create Colorful Anime Treasure with CAT!
The age of Pony is over, the time of the CAT has come! -- Lord of the Rings: The Return of the King (probably)
Update CAT IL v10 - (CAT_NAI_v2:0.6) x (IL_v2:0.4)
Update CAT IL v9 - (CAT_NAI_v2:1.0) x (AolaStyle:0.5)
Update CAT IL v8 - (CAT_NAI_v2:1.0) x (EnergyCAT:0.3)
Update CAT IL v7 - (CAT_NAI_v2:1.0) x (Unreleased_CosmicCuteStyle_Lora:0.6)
Update CAT IL v5 - (CAT_Illustrious_v4:0.5) x ((NAI x IL):0.5)
Update NAI v2 - This is (CAT_Illustrious_v4:0.5) x (NAI:0.5) so that its half and half. This way its closer to NAI's flexibility and knowledge than v1 but still maintains the high quality style of CAT.
Update NAI v1 - This is v4 merged with NAI in order to gain some of the Natural Language prompting abilities of NAI. This version should be compatible with NoobAI, Illustrious, and even a few Pony Loras. (There are several Pony Loras used in my showcase).
(rejected) Update v5 - Is more flexible with other styles and cleaner but loses the style of v4 so I'm going keep v4 as onsite generation. Here is a comparison of v5 and my baseline: https://civarchive.com/images/64724098
Description
masterpiece, best quality, very awa, amazing quality, 4k, very aesthetic, high resolution, ultra-detailed, absurdres, newest,
masterpiece, best quality, very aesthetic, high resolution, ultra-detailed, absurdres, newest,
FAQ
Comments (79)
I idea maybe creating Citron furry model
That's an interesting idea. Is there a good furry model that I can look at as a reference? What would be a good example for me to try to imitate?
CitronLegacy聽Normally I would tell you to make your own furry/scalie model from scratch, but if you don't want to force it, I can give you a few models that I can recommend.
Zannark聽All my checkpoints are one of style Loras merged onto a Checkpoint. (And/or multiple layers of that).
If that approach is good then I'll do that and train a really big furry lora for merging into CAT.
(Its my understanding that training a checkpoint from scratch requires a lot of expensive hardware so its not really an option for me).
CitronLegacy聽recomend this https://civitai.com/models/503815/nova-furry-xl
CitronLegacy聽Of course, as long as it is illustrative (and one more note, make sure the model stands out from other models)
Zannark聽Can you please let me know if what Galena CAT is missing that Nova Furry has? I'm trying to better understand the goal. Also how does it compare with CAT?
Here is a quick comparison https://imgur.com/AtDQwsO
Here are the images with the meta data: https://civitai.com/posts/20241988
CitronLegacy聽Let me put it this way: from the anime style perspective (if I wanted to do 2d furry) this anime style is best BUT when try creating 2,5d or 3d The other 2 models are the best choice (although knowing life, it needs some refinement) but I have to see how it will handle How to deal with dragons, naked scene, rare characters (like Bakugan wavern) and sex scene.
Baytheway Nova furry have also fotorealistic version maybe interesting
CitronLegacy聽a how long plan creating furry model
Zannark聽I'm not sure how long because I'm not sure what I'm trying to do yet. I spent 10 hours training a lora on 2500+ furry images but it didn't work well. If I can make a good lora then I can merge it into a checkpoint but I don't know how to make a lora that has all that furry knowledge like you are talking about. I think it would be better to focus on furry "style" so correct anthro anatomy. For this approach I think I need to find a few really good furry artists and focus on their style.
CitronLegacy聽when need help for creating model I know a few people who should be able to help you without any problem.
v7 definitely my favorite. Would love to get my hands on the Unreleased_CosmicCuteStyle_Lora that it鈥檚 based on
Agreed
Absolutely amazing. Works so well with my characters. Also pairs amazingly with Valentinedesigns Avarice LoRA. Seriously, give them a try together.
Hey, your description mentions "IL v10"
But where is it?
Oh shoot I forgot I left that comment there. I made and uploaded v10 but I'm not sure if its good enough so I didn't release it yet.
Recently I've been thinking of new checkpoint idea so I might release something soon.
For now I'll remove the comment because the current v10 will probably never see the light of day. lol
@CitronLegacy聽No biggie! 馃槉
I just had a funny realization:
Citron Anime Treasure = CAT
and in your description you put:
"Colorful Anime Treasure"
Colorful Anime Treasure = CAT
Two years ago I was trying to come up "good branding" so I went for something like a double entendre! CAT means both things. Lol maybe it wasn't that clever 馃槄
Oh I just realized I released the first version of the model on 2023-08-20! I missed this model's birthday lol
@CitronLegacy聽馃悎
Hey so a random thought I just had:
Since CAT IL V9 looks so good already, is there any real need for a V10? Just curious.
Thanks for saying that v9 is good!
You are right I basically achieved my goal with v9. With v9 and NoobAI v2 (this is more style neutral) I have almost everything I wanted to make so I don't really need more versions.
This is what v10 looks like: https://civitai.com/posts/21690421
v10 was supposed to be better at understanding "Natural Language" prompting because its based on Illustrious V2. Illustrious v2 is a bit unstable so it made CAT v10 a little unstable so I was unsure about it and didn't release it.
@CitronLegacy聽I meant to reply sooner but kept forgetting.
Judging from the few images you uploaded, I thinking IL V10 looks great! I especially love the white haired gal! 馃槉鈾ワ笍
The noobAIv2 version is awesome. I was using the illust v9 but changed to the noobAI version because the results are really awesome.
Is the model v-pred or based on the epsilon-pred one? if this is the epislon, do you have plans to update it or maybe in the future create a v-pred model?
Thank you!
Yeah I love NoobAI v2. I switch between v9 and NAI v2 depending on my goal.
Its Epsilon-Pred I'd like to make a v-pred version but I don't understand v-pred that well so I don't know how to tell if I make it correctly or not lol.
@CitronLegacy聽thanks for the reply.
Well, neither do I understand v-pred that much but I heard good things about it heheh I'm using mostly NoobAI-based v-pred models.
Well, if you ever make the CAT citron anime in v-pred, I'll be there to test it, if not I'm extremely happy with the epsilon-pred version. Keep the good working <3 and thanks
Update CAT IL v10 is releasing today (2025/09/28)! -- Please let me know what you think of it?
Feel free to give brutally honest feedback. If its a bad CAT we'll send it to the doghouse.
Pros: It makes really creative art.
Cons: Its hard to control.
Its merged with Illustrious v2 which means it "understands Natural Language."
I haven't learned the right way to prompt it yet. I don't always get good results but when I do the results are way better than the previous generation of Illustrious checkpoints.
Nice one! ive been waiting for the return of CAT. Can't wait to try this new version
@im_new_to_this聽Lol thanks! This version is chaos which is both good and bad. Personally I use v9 and NoobAI v2 most of the time but CAT v10 has hidden secrets!
@CitronLegacy聽which version will you keep available for on-site gen?
@im_new_to_this聽I'll keep v9 available for on-site gen. Unless the community votes to replace it with v10. Actually I regret that I took down v4 so I would hate to kick people off of v9.
v10 shows the pros/cons of Illustrious v2 so we'll see what the community thinks.
Someday I might be able to support more than just 1 checkpoint. For the last 2 months I've been slowly losing buzz because my checkpoints cost more than they earn. lol
@CitronLegacy聽V4 is so good. I love the vibrant and colourful anime style. I always saw v7 as the spiritual successor of V4, but I want to give V10 a try, so that's my vote 馃槂
@im_new_to_this聽Yeah v7 was an attempt at making v4 with bit of warmer tones but the community didn't use it very much lol.
I was considering bidding on both v9 and v10 but thats kind of expensive soooo I dont know lol.
@CitronLegacy聽can I help to bid on v10? Whats the benchmark to get a checkpoint online nowadays?
@im_new_to_this聽You can but you don't have to. Winning a bid is about 25k to 30k. (Imagine spending $30 USD a week ($120 a month) to keep a model alive.) The only reason why I'm able to keep CAT v9 alive is because all my loras earn enough buzz to cover the loss of Cat 9.
Anyway I'm kind of planning on paying 50k to 60k to get both v9 and v10 enabled this Thursday. If v10 doesn't get enough usages then v9 will be the only one that says alive in the future.
@CitronLegacy聽
I've used your checkpoints to create doujins / articles and got tipped generous amounts of yellow buzz for those. I don't know how else to use it, so 10k bid - here we go!
@im_new_to_this聽Thats very generous! Thank you very much!
I placed my bids so both v9 and v10 have about 25k each. If needed I'll put in the extra 5k or more to make sure they get enabled for onsite generation.
@im_new_to_this聽I checked out your articles and they good!
I love the Nami cover image on this one https://civitai.com/articles/12323/generating-images-on-a-budget and also appreciate all the references to CAT and I recognize that sexy art of Loid and Yor! Hmmm maybe I should have tried making a low steps CAT?
And this doujin is really funny https://civitai.com/articles/12846/nsfw-tohsaka-vs-captcha it perfectly captures Rin's personality. It reminds me of the scene in Carnival Phantasm where she is so bad at technology that she can't figure out how to use a DVD/DVR lol Poor Rin is not ready for the era of technology.
@CitronLegacy聽thanks Citron (and appreciate the generous tips)! 鈽猴笍
I need to update the article since some things have changed over time. I don't think it's worth spending time to create a low step checkpoint. There this lora that works surprisingly well for turning anything into low step: https://civitai.com/models/216190/lcmandturbomix-lora-only-12mb-8-step-sampling-effect-is-superior-to-using-lcm-or-turbo-alone
And yes! I watched that scene of Rin struggling with the blue ray player! It was shared with me when I explained that I needed a character who struggles with tech 馃槄
Let's get V10 online 馃槑
@im_new_to_this聽Thats really cool! I'll test that lora out!
Lol a true man of culture must have sent that Rin video to you.
V10 is online!! Thanks for the help getting it there!
@CitronLegacy it's been fun using it. The art style seems to jump around a bit (or maybe i'm not using the right set of quality tags), but it stabilises when using style loras. I've experimented with the natural language part and it does seem like the checkpoint is listening to it well!
@im_new_to_this聽Glad you like it! I'm enjoying it so thanks for convincing me to enable on-site generation!
I think its a bit unpredictable but that makes it more creative than other checkpoints. So far my favorite aspect is the abstract backgrounds or dynamic poses. I'm a noob at natural language so I had ChatGPT come up with some prompts for me lol.
I noticed that using a style tags or artists help with the instability.
Your post turned out really cute! https://civitai.com/posts/23160600
@CitronLegacy聽it's been great using your new checkpoint! The natural language bit opens up many doors for concepts that are not covered via danbooru tags
@im_new_to_this聽Please @ me if you post anything you like! I'm still exploring this checkpoint so I'd love to see what you make with it.
The heart/sparkles/vfx on this one are great https://civitai.com/images/105345982
And the cinematic scene in this one is awesome https://civitai.com/images/104183084
Its so cool seeing how much AI has evolved in the last year.
@CitronLegacy聽thanks for the feedback and appreciation!
I'm planning to do some kind of boss fight sequence very soon. I will tag you for sure!
So far my most ambitious sequence (and the one I'm most proud of) is this one which I did on another checkpoint. It's sfw.
Part 1: https://civitai.com/posts/22053748
Part 2: https://civitai.com/posts/22054079
I managed to include many of the loras that I often come back to. This includes one of yours. Do you recognize which one? 馃榿
@im_new_to_this聽Nice! My most useless Lora! 馃ぃ
This is a nice story! I need to try making stories again.
I really enjoyed both of these stories! I would be so cool to see the doujin world expand with new artists and faster content.
This is my favorite page Lol: https://civitai.com/images/99393159
And this one is really cute: https://civitai.com/images/99393667
.
A fight sequence would be hype. How do you plan to make the fighting poses look good? I always assume it will be hard to get the right angles. I have a dataset for fighting poses but I haven't figured out what to do with it yet because I don't think it will make a good lora: https://imgur.com/a/FQH89BQ .
@CitronLegacy聽thanks! And I definitely want to see some of your stories! Imo there are not enough folks out there that use this for visual storytelling.
Regarding the boss fight. I still have to think it through, but my thinking is that the main focus will be on the protagonist. The "boss" doesn't need to be in every frame, which makes things easier. I'm only using the on-site generator, so I'm limited in terms of the tools I can use. But I've been experimenting with img2img and I'm fairly confident that I can create a convincing illusion of a fight. For example you can create an image where the protagonist is dodging a "giant axe" thats causing the floor to crack. You can then use img2img to add the lora of the bad guy and add tag "giant hand" and "bad guy holding axe" or whatever... The bad guy won't be fully in frame for that image, but it will (hopefully) look like they are swinging an axe at our hero
POKE POKE POKEGIRLS 馃槏馃槏馃槏
Yep! If I've learned anything from Ash, its that you can't be a Pokemon master unless you have a few Pokegirls in your crew!
More of v7 is all I ever wanted. Release the Cosmic Cute style! :D
Oh yeah I remember you asked about Cosmic Cute. Sorry "A magician never reveals his secrets." If I give away all my tricks I won't have any all. 馃槣
If I start earning enough buzz I'll enable more checkpoints but right now I can't afford it. As it is I lost buzz every week for about the last 2 months just keeping CAT v9 alive.
@CitronLegacy聽that's sad. This is a great model!
My new goat model! It worked so well with my loras
Thank you!
You have a nice collection of Loras! I need to test them out.
Has inpainting halos problem (even with Soft Inpainting enabled) but it seems work fine
Does this model have a VAE baked in? Because when I used it I forgot to set a VAE but the colors did not come out unsaturated.
Oh! I didn't know it at the time but after doing some research I see that CAT v4 did have a VAE baked in!
Most of the CATs after v4 are based on v4 so they all inherited that VAE.
@CitronLegacy聽Okay! Also the version I used most recently was IL V10.
@AutMacro Hopefully v10 is working well for you!
Any chance of a ZIT version of this model?
Sorry for the delayed reply. I was hoping I could figure it out but no luck. The way I make checkpoints is merging my Loras into checkpoints but the workflow for it is taking too much VRAM so I dont have a way to make ZIT models right now. 馃槩
@CitronLegacy聽Well... dang. Also I looked into this right after reading your reply and it looks like you would need a workstation GPU with at least 48GB of VRAM for that which are stupidly expensive. At least according to Microsoft Copilot.
@AutMacro聽I was thinking I can rent a GPU server to do a lora + checkpoint merge real quick but I need to do a lot of prep to make sure I don't waste time on that rental. It would suck to be charged by the hour and then spend the entire hour just setting up the workspace/merge software. Also it means I cant experiment and make several different merges then pick the best.
I'm thinking about it but I'd say I need to learn more about ZIT first.
And anyway I think Illustrious is better for anime right now.
.
It would be so cool to have 48GB of VRAM. I make a lot of AI videos so that much vram would improve the quality a lot!
@CitronLegacy聽Sounds like a decent idea! Also, Just know that if you ever decide to buy such a GPU outright in the future that you are looking at thousands of dollars, maybe even 10s of thousands if you go full nuclear.
...Though you probably already knew that. 馃槄
@AutMacro聽Yeah 10k is insane lol. I guess if I was making money with loras and AI I could earn enough someday. Maybe?
I think a more likely option is combining two GPUs. Not all software supports it but there is some potential there. Also AMD sometimes has better price to vram ratios but I've never tried anything AI on AMD lol.
@CitronLegacy聽Combining two GPUs? Not a bad idea. Only one problem: You would need at least 2 RTX 4090s for that as each one has only 24GB of VRAM so you would still be spending thousands. Additionally, its not just RAM that has been going up in price because of the AI boom, SSDs and GPUs are being targeted too, especially the upper-end ones. Also, you are right: AMD GPUs might be worth considering.
I'm finding rejected v5 is amazing for a toon style
Nice! Thank you for saying that! I felt like it was a failure of a model so I'm glad you enjoy it.
Does it Support Artist Tags?
Yes it should. Its based on Illustrious and NoobAI.
Nice! As I saw some Illustrious models that doesn't support it. Or support it to a Certain degree. But thank you for your Answer. Looks good. Especially after I knew it'll satisfy my Obsession with Artist Tags nihehe
@Boredafk聽I haven't really tested with artist tags so hopefully it works well. This has NoobAI, Illustrious v0.1, and Illustrious v2 so its has a little bit of everything. Hopefully it works for the artists you are looking for!
I'm Impressed with your Model overall. Thank you for releasing! Almost everything I wanted is in this Noobv2 Model! Now let it Enter my Faves!
I been meaning to ask but any chance of an Anima version of this checkpoint?
Hi AutMacro, nice to hear from you again! Hope you've been doing well. 馃榿
Which version of this checkpoint would you want on Anima? Some people prefer v10 while other prefer v9, or even v7.
@CitronLegacy聽Hm, I like both V9 and V10. So why not both? 馃槢
@CitronLegacy聽Now if I had to pick one however, it would be V9.
Can you tell me up to which day this checkpoint supports the Danbooru database?
Hi thanks for asking!
This checkpoint is based on NoobAI v1.1 and IL v2 so it has a little bit of the dataset that was used in IL v2 but its most the 2024 Danbooru dataset.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.