Check out our website to see what we’ve been working on and explore our latest model!
→https://www.illustrious-xl.ai/
We’re excited to share that you can now generate images directly with our Illustrious XL models on our official site: [illustrious-xl.ai]. We’ve launched a full image generation platform featuring high-res outputs, natural language prompting, and custom presets — plus, several exclusive models you won’t find on any other hub. Explore our updated model tiers and naming here: [Model Series]. Need help getting started? Check out our generation user guide: [ILXL Image Generation User Guide].
Illustrious XL v1.0 – High-Resolution focused Illustration Generative Model
Overview
Illustrious XL v1.0 is a cutting-edge generative model developed by OnomaAI, built on the Stable Diffusion XL architecture, trained from our previous checkpoint, Illustrious XL v0.1, designed to produce stunning high-resolution images. It achieves an unprecedented native resolution of 1536×1536 within the SDXL framework, pushing beyond previous limits in detail and clarity. The model seamlessly blends natural language understanding with tag-based prompting (Danbooru style), allowing both descriptive sentences and specific tags for flexible, robust prompt handling. As a result, Illustrious XL v1.0 is a powerful foundation for creators seeking versatility in content generation without sacrificing fidelity.
Knowledge Cutoff
Illustrious XL v1.0 is trained in 2024-July, with knowledge up to 2024-June.
However, due to its native support of v0.1 LoRA, you might not have to worry about knowledge limits!
Key Features
1536×1536 Native Resolution:
First-of-its-kind Stable Diffusion XL model to natively support 1536px resolution. Generates images with exceptional detail, sharpness, and clarity at high resolutions that were previously unattainable in SDXL. It supports wide range of resolution between 512x512 to 1536x1536 - resolution such as 1248x1824 is capable without any high-resolution modifications.
NLP + Tag-Based Prompting:
Integrates advanced natural language processing with Danbooru tag-based prompts. Still, tag-based approaches are more concise and accurate - but this hybrid prompt system means you can use plain English descriptions or precise tags (or both) – the model understands and leverages each, giving you greater control and nuance in image generation.
Extensive Compatibility:
Fully compatible with a wide range of extensions and add-ons. LoRA, ControlNet modules, and other adaptation methods trained on Illustrious v0.1 work seamlessly with v1.0. You can mix and match enhancements to fine-tune style, pose, or composition without breaking compatibility.
Pretrained Base Model:
Illustrious XL v1.0 is provided as a pretrained base checkpoint without any fine-tuning on specific aesthetics or biases. This "raw" model offers a robust foundation for further training – ideal for creators who want to apply their own fine-tuning (e.g. LoRA training) to achieve particular art styles or specialized outputs. It's a flexible starting point for new innovations.
Future Plans
Higher Resolutions in v2/v3:
Upcoming versions (Illustrious XL v2 and v3) are planned to be released, supporting even larger resolutions, targeting up to 2k resolution and beyond. Expect even more detail and scaling capability, enabling ultra-crisp large-format images.
v-parameterization Models & Color Control:
A specialized v-parameterization variant is ready, which breakthroughs SD XL's previously considered limit, both academically and practically. Alongside this, future releases will focus on enhanced color control, giving users finer adjustment over color balance and consistency in generated images.
Continuous Improvement:
The Illustrious roadmap is ambitious – each iteration will integrate community feedback and the latest research. We will continue high-resolution research with breaking previously considered walls, with careful and sophisticated approaches.
Illustrious v0.1 (huggingface) | OnomaAI TooToon | Terms of use
Description
FAQ
Comments (45)
令人悲伤的是，1.0版本仍然有tag污染问题，onineko-chan污染onineko，kedama污染kedama milk等……
Actions speak louder than words. I'll believe it if/when it happens.
All they had to do was say something like "Hey, for feedback reasons we're going to keep the model online only for 2 weeks. Sorry for the inconvenience but this is important to us". People would have still grumbled about it, but only a little.
The bigger problem is that now Civitai allows this function, which is extremely concerning for the future.
@Giangiotto yeah, bad communication
@Hugs288 more like Mostly negative(532)
Onoma AI version 3.5 already done, those are blunt lies. hurrdurr community data collection analysis setting up future direction lol
This is damage control
@gugl653 it's not for this model serie, they have other ones in the works and other projects linked to this one.
They would like as much feedback per version that they made.
I don't understand this community... V0.1 was a leap from everything else before (not counting NAI). Without it we would still be using score_X and having to train a LoRa for every character and every style.
I can assure you that this was the plan from the start, angel is one of the dev, he wasn't in charge of this communication. Because of his proximity with the touhouAI community we have updates and sneak peaks of the progress they are doing.
They don't deserve this hate..
@Poochilli That's cool, but I thought that 0.1 was also actually leaked. They even stated that directly in some post.
@Volnovik It was leaked days before the official release was due. Many people already had access like a week earlier to test, me included, then someone that had access leaked it on 4chan. It wasn't an online demo, we all had the weights locally.
Purchased EA and made comparison so you dont have to.
TLDR - I'm unimpressed so far.
All gens in Forge at 1024x1536 which is adwertised to be upside of this particular model, so others are at disadvantage from the start. No loras, no tricks, no cherrypicking:
https://civitai.com/posts/12906104
The only thing is that I used sampler and latent modifier config that works for v-pred. But it seems to work best for this model also.
Tomorrow will do one with wonkier resolution stated in description.
Upon gathering data for comparsion I found an interisting thing. Despite claims in model description, in their technical paper only 1.1 and further models are stated as being trained with NLP tags, not 1.0. We'll se how their "release schedule" works, currently from official civitai newsletter it looks like 1.1 will not be available for download.
That's strange considering there's a guide here based on 0.1 that said natural language prompts work great alongside tag prompts. From my experience, natural language + tag prompts can be hit or miss, even on merges/checkpoints.
@Wallace_Tanaka My tests has shown that capability is there. Best way to use it is to first go with tags, then and descriptive sentences like here https://civitai.com/images/62139514 This greatly enchances the image, but Illu1.0 is sometimes hillariously bad, especially in nsfw.
Not sure that someone still cares, but made a legthy article comparing this model in extreme resolution against other models that are not supposed to work under such conditions.
I was actually really excited to try this locally, but... No matter what I try - no matter the scheduler, sampler, steps, CFG, prompting - I get just... Really bad gens. Like, gens that look really, really bad. I've tried the settings from your example images, but I get nothing near that quality. I'm feeling kind of disappointed because this cost $10 to try aha...
Try settings I used in my article
Personally, I think the first thing OP need to improve is not the model's ability to generate higher resolution images, but to overcome or improve the problem of character features contaminating each other, which is a very serious problem. And there are many ways to improve the resolution of the generated images at the moment , and it doesn't need the model itself to improve this ability in particular.
Hell, it would have been better, if they didnt release this. Now 0.1 loras dont work on 1.0 based models, even with correct name.
According to the Huggingface;
"Note: The model does not have any default style. This is intended behavior for the base model."
So it's going to look generic without some rigging.
I'll play with it and see what I can come up with.
What sampler and cfg did you try?
@P_Universe I'm converting it to VPRED but it's a complex one.
@AbstractPhila damn that's awesome
@P_Universe The devs did a partial conversion but it's imperfect and it seems like they recognized it as imperfect as well by calling it failed.
I'll be using my own ground work to fix it.
They ruined character likeness. Lora makers fix it, then they release new version, which wont work again.
Good job.
Do realistic finetuners leave this crap, and focus on Pony7.
recommended settings?
Illustrious v3.5 v-pred has already reached 100% stardust level and even higher. Their website literally says "We promise to open model weights instantly when reaching a certain stardust level (The stardust % can go above 100%)". If they were lying about this, it could be considered a scam because people literally paid for the model release. Hopefully, the model developers will keep their word and release the model weights soon. If the model is really as good as they say, I think people will be willing to buy their future models. https://www.illustrious-xl.ai/sponsor
i already had my pitchfork in-hand so i went to the linked sponsor page to see just how mad i should get. at first i thought the way they set up the tiers was dumb because it'd allow people to just donate to the most recent model and it'd be released for everyone without the need to fund the others. BUT. upon further inspection there seem to be two different sets of incentives. if i'm understanding it correctly you have to donate under that particular model to get access to it. right? they have a meter at the very top of the page that says "Global Stardust Progress Meter" with a mouse-over that says it's to measure how far away they are for open-sourcing a particular model. the vpred 3.5 is the last amount at 53 million (it says we're at 1.2 million now.) i assume i'm missing something because i don't follow these things very closely so idk wth has transpired with this. what's going on? am i missing something? did they change the terms halfway through? pls halp, no bulli
V2 is available at Yodayo:
Illustrious XL v3.5 vpred hit 100% funding. You promised to release the model files instantly at a stardust level (which can exceed 100%). Will you release them or cast "take the money and run"? https://www.illustrious-xl.ai/model/8
329.220 "Stardust" raised at this time. Since they sell "Stardust" for one dollar a 100, does this basically mean they scammed people out of over 3 grand? That's not peanuts and getting into the realm of legitimate fraud.
And then they also change the overview page to what is basically an advertisement for their own generator? (Just like PDXL did.) At this point, Civitai might want to make some new rules about what a model owner can link to on his overview page, since these are becoming nothing more than an elaborate way to spam.
You have deceived people by claiming you would release it upon achieving 100%. Congratulations on becoming a true fraud https://www.illustrious-xl.ai/model/8
1.1 is available for free download right here: https://civitai.com/models/1252206/illustrious-xl-11
There is no deception, there is no fraud.
Well compared to the horrible garbage of your "2.0" this checkpoint does the job . It's got strengths and weaknesses with highlights to it that make it worth using UNLIKE THAT OTHER TRASH
why you gotta be so mean for no reason
entitled cringe, vomit.
Can someone explain this drama to me? Did 1 person release Illustrious and then an organization bought it or was it always this organization behind every model?
Only 15% remains until Illustrious XL 2.0 hits the next milestone. Sponsor it or not, it’s up to you. But if you want the 2.0 version, give it a think. https://www.illustrious-xl.ai/sponsor
They launched an image generation service on their site, but it’s more expensive than NAI. There’s no reason to use it. They don’t even communicate. NAI has great communication, frequent updates, and is currently testing NAI 4.5, which seems to be launching soon. Don’t waste a dime on their lousy image generation—subscribe to NAI or use WebUI and ComfyUI instead.
Could we expect an inpainting version?
It's not SD 1.5.
You inpaint with the same model. Illustrious excels at inpainting. Just try...
>224 hidden comments
Lmao
"This model is not currently available for LoRA training - please select a different model." - why??
stop making sh*t
Details
Files
illustriousXL10_v10.safetensors
Mirrors
Illustrious-XL-v1.0.safetensors
Illustrious-XL-v1.0.safetensors
Illustrious-XL-v1.0.safetensors
Illustrious-XL-v1.0.safetensors
ilxl10ForLora_v10.safetensors
illustriousXL10_v10.safetensors
illustriousXL10_v10.safetensors
Illustrious-XL-v1.0.safetensors
ilxl10ForLora_v10.safetensors
illustriousXL10_v10.safetensors
illustriousXL10_v10.safetensors
illustriousXL10_v10.safetensors
Illustrious-XL-v1.0.safetensors
illustriousXL10_v10.safetensors
Illustrious-XL-v1.0.safetensors
Illustrious-XL-v1.0.safetensors
OnomaAIResearch_Illustrious-XL-v1.0.safetensors
OnomaAIResearch_Illustrious-XL-v1.0.safetensors
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.