🕊️ This model uses illustrious XL as a base instead of Pony XL. This comes with several advantages:
A lot more characters and concepts, artist tags, and increased accuracy
More varied and detailed backgrounds (pony was trained without backgrounds)
Multiple characters in the same image
Sadly this also means most Pony Loras won't work.
Positive prompt: masterpiece, best quality
Negative prompt: worst quality, low quality, lowres, jpeg artifacts, bad anatomy, bad hands, watermark
CFG Scale: 6, or 5 for lower contrast
Additional Information
I didn't use any Loras or regional prompter in the previews.
The style is different compared to Honey Mix. Flat shading was desired. The model showcases high visual fidelity (specially with hires fix). Contrast is even higher than Honey Mix, and it may be more or less vibrant depending on the image. Some people may think it's too much, but I like it.
Hands are very good, but AI is AI, so it won't be perfect every time.
NSFW is also great, all your kinkiest tags are probably built-in, but I didn't focus on it in the previews.
ADetailer is recommended to make faces more detailed, like always.
This model is a merge of WAI-NSFW 6.0, Spatter 2DV2, PaSanctuary 2.5, and my own little special sauce. Thanks @PotatCat for the help.
Check out my new website, a comfy place where experienced AI artists and model creators can share their work: https://arcenciel.io
Description
FAQ
Comments (34)
This is highly impressive for character generation without a LORA, that is almost unbelievable.
While trying to replicate the prompts by OP with Krita AI, sometimes the images are too over-exposed or saturated. Removing most of the negative prompts and reducing guidance steps solved it for me.
Thanks!!!
Hola compa, que pasa con este nuevo modelo, veo que "illustrious" esta en ascenso, aun no lo pruebo, tan bueno es?
:FubukiSway:
(～￣▽￣)～
vae contained?
I think so yes, though you can also use your own
Amazing model! I love it!
We ball?
Great model. You do notice that the characters with more art were easier to prompt than characters with less art (or newer). Still a good understanding of concepts.
EDIT: At this point this is still my Nr.1 when it comes to Anime/vtuber generations.
Fantastic model! What clip skip do you recommend when generating images with this?
That would be clip skip 2
@holostrawberry Thanks! 😊
Amazing Model. Love this style! I'd like to use it for my hololive lora!
Go ahead!
Beautiful. still learning to fine-tune, though.
why i cant use loras? i tried a lot but loras still doesn't work
Any loras you use need to be made for illustrious, not pony
@holostrawberry My Pony lora's are actually working almost perfectly. As well as the XL lora's, interestingly enough.
fantastic merge, demonstrates high fidelity and quality while retaining a natural anime aesthetic, maintaining the nice sharp colour contrast that makes illustrious based models great providing nice dark blacks and vibrant reds.
This model is one of the best that I have been able to use consistently with the lora Hyper4 at 6 steps.
That's very stable.
Man, i love this checkpoint, but i cant seem to merge it with others, keep getting "Error merging checkpoints: argument 'metadata': 'dict' object cannot be converted to 'PyString'"
I would recommend merging with ComfyUI
@holostrawberry do i need a specific module download or something, or is it native out the box? i dont use comfy much.
@SomebodyOnceToldMeh you'd have to find a module/workflow made by someone else. I don't use comfy much either but I used it to merge this checkpoint. A friend gave me a workflow
Great model, I think it is the best model in Illustrious. I would like to see a newer version of the model (merged with the latest version of WAI-NSFW) if possible.
Any chance to update this to Illustrous V1 when its released?
It's been a while and I think most people agree illustrious 1.0 isn't great, but I might make a noobai checkpoint instead
@holostrawberry I agree V1 sucks... But A Noobai Mix is wholeheartedly welcome. That would be great!
@holostrawberry any news on this?
This checkpoint is Top Tier. I really liked standalone capability of it. Its Really easy to get detailed scenerys from it, very consistent in work. Depending on task might perform better then WAI's. For flat color images its straight up best.
really love the artstyle but small details tend to have more artifacts or inconsistencies compared to some other models
You say that this doesn't work with PonyXL LoRAs because it's an Illustrious model? Oversimplification. PonyXL LoRAs, from my and others' experiences, work with Illustrious checkpoints!
Why it's doesn't work on online generator, I really loved this checkpoint ):
It's too expensive to keep it in the generator
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.