AniMixV9XL AnimeTV Mix.
AnimeTV/Screencap Fine-tuned Model's Based on AutismMixXL & PonyV6. 🔥
All Examples image are RAW. (TxT2Img + Hires.Fix & ADetailer).
UPDATE: Check ''MegaMixXLV3'' My new Checkpoint AnimeTV Fusion/Screencap (Based on Illustrious/NoobAI V1.1) 🔥 AVAILABLE FOR ALL GEN RIGHT NOW!
Power of this model's
Have an AnimeTV/Screencap Style. (Obs).
Capable to change a No-Character/Game Character into AnimeTV Style. (It's also Enhance/Improve some of the AnimeTV/Screencap Like Lora's into more AnimeTV).
Capable of inputting some character's name without lora's (Example: nami (one piece)/raiden shogun Booru Tags.
Pretty Good at SFW/NFSW+.
Anatomy is ''Ok'' in RAW, Easy also to Fix it with Inpainting.
Got a Big Buffer = ''Anime Screencap'' Putting it in Positive prompts = Even more Anime Screencap Style. (I Recommend).
Suggested settings
Sampler: Euler A. (LCM+ Can Work too, just lower the steps and CFG).
Steps: 20-28.
CFG: 4.5-7. (5.5 CFG = Best of both world in my Opinion).
A Good Resolution: 768x1192/832x1216.
ADetailer & Hires.Fix: Is Recommended.
Trigger Word: Anime Screencap Prompt = Buffer, (It's work without, But it's much Better with it if you want a Full AnimeTV Style).
At Beginning: score_7_up, BREAK source_anime (Combine it with Anime Screencap for Full AnimeTV Power).
Don't use negative prompts or Embedding that will deterior the Style. ([Important] Check Examples images for Negative prompts examples).
VAE: Click Here - (Better in FP16).
For Lora's/Model's Creator: It's might be a good idea to train on this model, if your character is a Full AnimeTV and your datasets is it too. (Need to test).
Otherwise i would recommend all the time to Train on the Official/PonyV6 for Full Compatibility.
This Model's have 3/4+ Of my personal Lora's trained specifically for this AnimeTV Style. (Also Added AnimeTV/Anime Background Lora's).
For the Little Story: I created it for the simple reason of: Since PonyV6/AutismMixXL was Lacking Highly in AnimeTV/Screencap Coloring without a Lora's and Even with them sometime the Style was not Consistent.
It was a Challenge also to have the advantage of PonyV6/AutismMixXL (Anatomy, Booru Tags, ect) And to work full without a Trigger Word (Yes because Pony is Not very good without, It's not like SD 1.5/SDXL, you need to hard train a Lora/Put more Weights, This model's taked me Many Die & Retry to create ahah), Now he work without AND have a Buffer with ''Anime Screencap'' Prompts.
Also i'm pretty Satisfied to be the First RAW AnimeTV/Screencap Model's on PonyV6.
More Update's coming it in the Future.
PS: Don't hesitate to post your Artwork/Image, I like to see what Other People's create with My Model's. (It's also help to popularize AniMixV9XL).
PS2: If you merge it into a Model's, @ me on your Model's, or Talk to me in private.
Description
NEW VER AniMixV9XL-V2.5/REV - Update (What's New?):
Well, This Version of AniMixV9XL-V2.5/REV Have many changes in it:
I say in past comments that i will reduce pixelization, I finally got a good balance after many days, Now AniMixV9XL-V2.5/REV have 40% Less pixelization than before at minimum. (40-50%), is now on pair with Other good's model's. (Tested many of them 3D/2D).
A little better quality output, More Stable & Clean in general, Less artefact since Pixelization is also reduced.
Working even better with Lora's, Specially those who are Full AnimeTV, he's enhancing even more the Style. (Like in some my examples images, where some of my Full AnimeTV Lora's are in it)
Anatomy, is a little bit improved specially in Close-up/MidShot. (Specially Barefoot Ahah).
A little bit better also at Background/Particles.
Well, After many days of balancing the model's & All i'm clearly satisfied about the result, He's doing what you want from a RAW AnimeTV/Screencap Model's with some Good Background/Particles & Anatomy's that other's Model's have.
FAQ
Comments (22)
Nice work, a very clean simple version of pony. Great for quickly creating flat anime style images! Save on Lora bleeding / pollution
Ahah thanks, that's was my goal to have remove some of the Pixelization (Even if removing full is impossible for a Style that Pony Model's don't know), More clean/sharp, Keeping a Good Background/Particles & A Little Better Anatomy. (Did write all the Info's on About this Version).
Any ideas how to stop getting simple backgrounds? every wall/ceiling has a single color and even adding them to negative doesn't change the outcome
Mhh, the V2.5/REV is Normally boosted in Background without sacrificing Quality/Anatomy/Pixelization, Tho Background was always a problem in Pony's Checkpoint.
The things you can do for augmenting it: Putting in Negative Prompts > (simple background:1.3) & the other things you don't want. (Ik you already did it).
But in my opinion the best way is to describe in positive prompts the most you can, Tho if it's doesn't work ''Good'' You can just go with a Backgrounds Lora's and normally all the things are fixed. 👍
@KXRO Thanks. Any background loras that work good with this model?
@extendou I can't really tell you some names because i don't use it when i'm creating my Artwork ahah, But generally All Anime Background are Ok's, The best to do is just to test them and pick the one you prefer. 👍
Hello, can you advise the best lora training config with this model as well?
Hello, if you mean to train with this Checkpoint, You can go with the default that you got on Civitai Trainer (Or just pick MetaData from Someone else/Also Adjust Dependings on what you want Character's/Style's/Concept's), But if i'm 100% Honest, I think the best is just to train on the Base PonyV6 for Full Compatibility. (It's work also to train on this Checkpoint tho, Some Tester have already tested it & Myself too, But i got back to PonyV6 Base/Compatibility Wise).
Yes, ok I can try the PonyV6 then. Can you help to check if below link is suitable training data to your Check point? LoRA Easy Training (dev derrian-distro) for SDXL Pony (with 3070RTX 8GB of VRAM) (August 2024) | Civitai
@N_A_L It's probably ok, Specially since it's on Easy Lora Trainer so you can't go wrong with it (I don't use it tho, But i know many use it and it's good since They have CAME/REX | Optimizer/Scheduler + Many Other Settings).
You thinking about making one for Illustrious? Love this model personally
Yea, Like i teased in Description, I already finished the Checkpoint, I got to the point where i Created 3 Version of It ahah, So you will already be in V3 ''Stable'', He will be a little bit Different, But have a Higher Potential + Better at Anime/AnimeTV In General, Far more Official / (CF My Recent Artwork on Pixiv 🔥), I will release it in Fews days 🔥 (Also created some Lora's for Illustrious too).
And for the Other who want a update to AniMixV9XL V2.5/REV (So in Pony, I will probably wait PonyV7).
Currently training a test LoRA using your checkpoint. The idea is not to make a LoRA to publish but to use this LoRA trained on your checkpoint to generate consistent anime images to expand an existing dataset which will then train a LoRA on base Pony
Love this model, but it has some problem getting right color eyes when used with loras, is there any fix to that ??
If you use a Character's Lora's, Normally (If the creator pruned the Eyes Colors) You don't need to prompts for it, You will have it 99% of time.
If he didn't pruned it, Just prompt it EX: Red eyes, Yellow Eyes, (Ect) in Prompts.
If it's a Lora's for gettings specific eyes (Example one i did for Daki Eyes - Demon Slayer/KNY), You will need to use ADetailer/Inpaint (More Easy to Inpaint).
And if you talk about Eyes who are Heterochromia (2 Eyes of Different Colors), Normally you have it without too much problem, But AI can be RNG (It's like Fingers/Toes), Sometimes it's place it to the left (Even if the Original was Right Eyes), But it's for all Checkpoint the same problems for this, So DW.
@KXRO i was having problems with momo's eyes from dandadan, momo has brown/red eyes, but i was getting full brown all the time but i guess it can just be the lora 😂 anywayss, keep the good work ^^ !! this model is nuts 😋
@waitercolor Oh, Yea probably from the Lora's then ahah, Her eyes is a little bit confusing in terms of Colors, So perhaps the WD1.4 Tagger putted too much Brown eyes on his tags in the Datasets ahah, Speaking about Ayase Momo a Friend created a Special Lora's on her. For my New Checkpoint (Based on Illustrious who will come), The Result was Very good !
Thx for the Kind Message, If you like AniMixV9XL V2.5/REV (Or Old VER), You will even more Like the New Checkpoint on Illustrious ahah, He's really goated if you know how to use it - (Far more Official on AnimeTV & Can Juggle Between Different Anime Coloring Style Easily).
Then for a Next VER on AniMixV9XL V2.5/REV, I'll probably wait for PonyV7 or some things others' ahah.
@KXRO Dang that sounds great ^^ will try it for sure 😊
@KXRO Is there any news on IL new model ?? would love to try that one tbh 😂😇
@waitercolor Yep, i did have updated my Profile text: ''Update 27/03/25'': (Illustrious Anime Checkpoint Finished (Based on NoobAI V1.1)/Also An SDXL Animagine V4.0 More ''Anime'' Version, Will Be Released Soon too (In work rn)- All are Finetuned/Trained by Me, Both have HIGH Potential for AI-Generated & AI-Assisted/Manual Artwork). Stay tune. 🔥
I was a little bit out of AI these times, But will continue to post semi-often content ! (And Lora's too, I have some to post ahah)
Thank you! This is by far the best Flat 2D Anime screencap model out. Really the one true one. As some has previously pointed out, the area that could use some TLC would be the backgrounds. Could never quite get those up to the same quality. Also, is this style possible to port to Flux? Base Flux doesn't seem to really care for real anime, doesn't seem like it was trained on a lot of it.
Best model! Thx u!
Details
Files
Available On (7 platforms)
Same model published on other platforms. May have additional downloads or version variants.