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].
Before we dive into the 2.0 model description, we owe you all an apology for the radio silence. We weren’t slacking off - we were cooking up something big. Over the past few weeks, we’ve been not only refining Illustrious but also building our very own site. Now, we have a home where you can get the latest insights on Illustrious, our training techniques, updates, and more.
From now on, all updates will be posted directly on our site!
For the v2.0 model’s technical details, check out: Illustrious XL Tech Blog
Stay in the loop with our latest releases and updates at: illustrious-xl.ai
Oh, and one more thing - we’re gearing up to showcase our models’ high-res capabilities and enhanced prompt applications soon, so keep an eye out in Q2 2025!
Illustrious XL 2.0: Increased robustness for better fine-tuning capabilities
In this version, the dataset has increased drastically, including major focus on animations & natural language focus – with updated dataset, until 2024-08. The model has performed best finetuning stability in our internal tests – with highest TrueSkill rating(among v1-v2) in external demo evaluation.
We want to note that we consider the compatibility & finetuning capability the most important feature of our baseline model. The model itself is not trained with aesthetic set / and we tried our best to debias the potential biases in model, for better overall performance.
Thus, we release ‘untuned’ base version, which should work as better merging / training bases. With release of the models, we also showcase example LoRA adapters, which is widely compatible across versions with recent character knowledges.
Natural Language Handling
The model itself is highly compatible with natural language sentences – it is far more
robust, it is less likely to generate multiple views or nonsense outputs.Generate a highly detailed anime-style illustration of a young man floating serenely
above a sprawling, futuristic cityscape. The boy has dark, messy hair and piercing blue
eyes. He's wearing a long, flowing white coat over dark, streamlined clothing – think a
mix of traditional Japanese garments and futuristic techwear. His expression is calm
and confident, almost detached. He is surrounded by a faint, glowing aura of light,
possibly blue or white.
Below him is a vast sci-fi city, filled with towering skyscrapers, holographic advertisements, and flying vehicles. The city should have a vibrant color palette – neon blues, purples, and pinks contrasting with darker metallic structures. There should be a sense of depth and scale, with buildings receding into the distance. The overall atmosphere should be epic and awe-inspiring, suggesting a powerful and mysterious character overlooking a technologically advanced world. Focus on dynamic lighting and detailed textures to create a visually stunning image. persona 3, masterpiece
High & Low Resolution supports
V2.0 supports 512~1536 resolutions. It is trained max 1:10 (or, 10:1) aspect ratio, with width / height multiples of 32.analogous colors, scenery, forest, masterpiece, black theme, dark, no lighting, night,
black background
Below is the 2x2 grid generation in 512x512 resolution:
1girl, solo, natsuiro matsuri, selfie, saturated, masterpiece, jirai kei, naughty smile,
dark, very dark, black theme, masterpiece, black, high contrast, sketch, monochrome
Negative prompt: worst quality, bad quality
Steps: 28, Sampler: DPM++ 2S Ancestral CFG++, Schedule type: Normal, CFG scale:
2.5, Seed: 1934035447, Size: 1824x1248 (NOTE CFG scale should be 6.5 in recent
implementations)
The models are capable for extreme high resolution generation more stably, with
higher denoising strength, which is important for stable detail generations. Below is an
example with 8MP img2img process:
We are also analyzing how to tweak the sampler schedule for better detail generations
– we will try our best to showcase the results.
Finally, we note the definite future plan for the model releases. We have decided to make model release sequentially, in three stage manner:
A. Exclusive
B. Early access
C. Open source
We will also continue to train & release the future model series, including DiT based
models.
Safety Guidelines
The model can be controlled by optional “rating” levels, described as explicit / questionable / sensitive / general
The alternative tokens, such as “adult content”, “r-18”, “nsfw”, “explicit material”, “sfw”, “safe for work” works for moderation.
Generative models can sometimes result biased, unintended outcomes. Users are required to generate and control the materials in responsible manner.
If you appreciate Illustrious’s evolution and are anticipating upcoming model releases, rlease consider leaving a like or comment. Your feedback is invaluable, and we deeply
appreciate your continued interest and support. Thanks!
Description
FAQ
Comments (176)
3.5-2.0 = Hey you owe use 1.5 more!
And... weights?
no local then?
You can tell the example gens were done locally since the prompts are all hidden lmao.
A. Exclusive
B. Early access
C. Open source
(Sounds like we will get the model eventually?)
someday. 300 years from now.
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
y'all got scammed lmao, not even Astralite hit this low
Snake oil salesman, I need your STRONGEST snake oil!
No bullshit, the community's problem with you guys is that you are never straightforward with your model releases. If you need $100K to release the model just say it from the start. Don't be like "Give us $3K, and the level can exceed 100%, but if it hits 100%, we may or may not release it instantly at 100%, but the level can go higher than 100%, which at some point between 100% and infinity we will release it instantly". Like what the fuck even is that shit lmfao.
Somebody threw a big money for 3.5. They didnt expect this, thatswhy they now think, they wont release it:)))) Just after exlusive and plus early access. Freakin scammers.
Congrats on destroying your reputation
So you just stole people's money and you're not going to release 3.5 v-pred weights now as promised? That's a crime, you know? Proof: https://www.illustrious-xl.ai/sponsor
Just reported them on CivitAi and TensorArt for stealing people's money. I don't know if this will work, probably not. But you can do it too so there are more of us. As far as I understand, reputation on TensorArt is more important to them than on CivitAi https://tensor.art/models/840364313702172787/Illustrious-XL-v2.0-Pro(Aesthetic)-TensorArt-Exclusive
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
I think your team makes good models, but your approach to releases is seriously flawed. You make funding goals with a progress bar leading to 100%, and then say "actually it needs MORE than 100%!". These are finetunes of SDXL, a model released almost two years ago. They are hardly on the frontier of prompt comprehension and complexity. If you need more funds then be honest with how much you need. $3k is obviously not enough to fund a full finetune, so don't put it there on the website to bait a quick buck.
Just release what you have regarding SDXL and form a plan to move forward with newer architectures. The community would support you if you were earnest about research, but gating off models trained over a year ago when NoobAI already exists is just silly.
The model understands "natural language" (prompts), but the development team does not understand "natural language" (user feedback).
Korean law, dear scammers.
1. Fraud
Article 347 states the following punishment for “regular” fraud. “A person who defrauds another, thereby taking property or obtaining pecuniary advantage from another, shall be punished by imprisonment for not more than ten years or by a fine not exceeding twenty million won.”
In other words, if a fraudulent act is committed, as described previously, the criminal might face up to 10 years in prison or could receive a heavy fine.
Even if a company steals your money, the laws and courts in South Korea are very lenient on scammers. Never waste your money. You will never get your money back.
In Korea, even if someone commits fraud worth tens of millions of dollars, they only serve about one or two years in prison. In some cases, they may not go to prison at all if they receive a suspended sentence.
They must convict before the chinese court:)))
Are you saying that starting from version 1.0, each version requires reaching a $3,000 crowdfunding goal before downloads can be sequentially unlocked? However, now that the crowdfunding progress bar is full, you've taken no action. I believe you need to provide a proper explanation and appropriate compensation.
well guys.....i think there is a misconstrue....
3.5v-pred version donate comes to 100%,but 2.0 and 1.1 are not in 100%...just check the website and slip down...
onomaai did a stupid settings that they put donate plans of all versions same time,their goal is that you guys to donate for every version,but 3.5v-pred which is the newest version is on the top,so everyone donate for this and didnt see what's below....hahahahaha
btw,3000$ for one version is a reasonable price,i think.
lol
Why would we donate for older versions? They are late to the party.
@mathewwatson915 without donate for older versions how can we get newest one? its old but it cost money too.
so in my opinion this is a setting mistake,lol
@NunFish They should have been released, when they were ready. 0.1 was leaked and they didnt react anything, till it became popular (because of community, not because of them). Now they want money for every outdated release, which are not even compatible with community loras(0.1).
Thats not how its working. If they released the 3.5 now, and later the newer versions with early access and crowdfunding, than it would be ok.
@mathewwatson915
you didnt pay a single cent. you are owed nothing
@mathewwatson915 i think 3.5v-pred is high probability not finished......
@NunFish They are just slowing down the community. V-pred wont work with previous loras anyway.
So these retards slowly release non V-pred versions, to fuck with lora makers. Very bright and shortsighted.
@mathewwatson915 im speechless. you are right,have a good time:)
@NunFish You too
@NunFish Maybe not be radio silent for over half a year if you want your models to still be relevant enough for a price tag?
@NunFish what a stupid take holy shit
It's like a company needing to pay all the expenses of an employee because he had to eat and had bills from birth till the moment he got the job, what in the fuk?
We donate for what we want to get now, you pay for what you want now, not for unrelated specifics outside of what you are promised or contracted.
@undef01
>we
how much did you donate? :)
@undef01 I have no intention of arguing,but this is absolutely a misconstrue,so onamaai needs to take the responsibilities for this.
i just come to clear there is a misconstrue,its good to stop a anger because anger cant solve problem,but if the anger cant stop,please dont mad at me:(
I mean I'd donate but if the result is that they release a generation only checkpoint then they can go screw. Their business model at this point is so convoluted they may as well make it a mobile gacha game. I miss when people would say here is a product. Buy it.
The web front-end designer at your team has placed the old model and new model together (even though the new model is actually positioned higher up, while the old one requires scrolling down to see), so no one would sponsor the old one.
You are a liar who has let down everyone's beautiful feelings and expectations.
Dataset is pretty old. Don't know TYBW Bleach characters.
i want illustrious 2.0XL not 3.5v-pred or whatever who the fuck cares about v-pred people???
ilxl vpred If you haven't paid, be quiet. Or sponsor 2.0 and fill it up to 100%.
me
im so pissed at people paying for v-pred 3.5 they should focus on eps versions! they are far more easy to train and develop
Illustrious Tech Blog - AI Research & Model Development
Some communication from Illustrious Blog.
chat why is this recommending a sampler/schedule combo we literally cannot use in civitai and then also not letting us dl the model
Again, no checkpoint available......
It's a pity that reality doesn't put scammers in jail where they belong; let's not feed the scammers
this is onomaai disrespect button 🐷
HUGE MISLEAD! To fix your reputation, just release v2.0 model file already before it gets worse XDXD 👀
They even put the creators name to sponsor page.
What a piece of garbage "company" (2 fat bastards in an office chair).
And they want half million bucks. Hahaahaha
Just watching the fire burn.
Hunyuan and Wan are a godsend to fool with until the REAL open source Flux/SD3 tier anime model comes out.
Or just stick with Noob. I'm not sure what this series of models even thinks it's accomplishing that hasn't already been done?
Noob can gen over 1024x1024 too...
No perma hate. I'll warm up to this if they can prove it isn't just a train wreck. Feels go out to those who wasted money on a "cannot download" model.
have you ever made commercial stuff with NoobAI based model ?
if yes... you have been screwed over then
@Le_Fourbe I've seen that brought up before. NoobAI is a finetune of Illustrious, which renders their license invalid. Even if not, it doesn't carry over to lora generations since those are personal creations. Finally, to be blunt about the AI art space as a whole... there isn't much one can do if a image is generated with enough variance from the model (through lora training), with metadata removed. This is entirely an honor system until some form of enforceable AI copyright system is established.
So using a undertrained experimental model that seems more interested in the niche breakthroughs, versus a fully developed well trained model with a future AGI themed update, sounds like a waste of time.
Also, it is likely the entire illustrious development is planned to end up as a pay per generation model like NovelAI (they are planning their Stardust site to be a model host if you can't tell). So it's pretty clear the best versions will be kept there even after all the "freemium" models are purchased. Unless Noob eventually does the same, Noob is still the way to go if you want to create your content freely without relying on a service and the issues those present.
I admit many of these may be assumptions, but until I see these disproved I will keep my stance.
How can I train a Lora with this checkpoint?
You'll have to Train on Tensor since Training probably only available there until this gets open sourced
@TatsuyaAI
TensorArt scammed me enough already. I will wait.
@alexds9 try shakker then
After the community strongly condemned your deceptive actions， you opted to surreptitiously add a Global Stardust Progress Meter on the website. Even with whatever technical breakthroughs your model may possess， this operational mishap will undoubtedly remain a lasting stain on your teams record. Moreover, I believe your premium-priced Advanced Checkpoint at $530,000 is likely to struggle to achieve its sales targets given the current climate of diminished credibility.
I’ve got a small Stardust investment and dig your model! Let’s wait for the new version. Stick to SDXL or SD1.5 for training high-consumption models will flop like unused 5G networks. RTX4090s are rare now, so no one’s chasing pricey models. SD1.5 is the smart choice: low-consumption, efficient, easy, widely available, and super practical! It fits all setups, even basic gaming rigs, ignoring graphics card price gaps. Prioritize the masses’ needs, and the cash will naturally fill your pockets.
1.5 missed the boat. They cant go back in time in AI field.
Due to the RTX4090’s crazy high price and rarity, AI might slide back to old, outdated models yep, that could totally happen, no joke!
@kz998653597 Allright, but XL is usable with old cards, only needs 4-6gb vram.
@davidphillips Also, if the generation is slow, you can use Turbo lora. Stability's Turbo lora also allows you to reduce the image resolution to 512, which also speeds up image generation on older GPUs
SD1.5 crumble down after 10 words really. the dynamic color range is dull, the details being captured require a LOT of additionnal tweak and fix.
you should only run 1.5 based model when you have a weak GPU or need to iterate draft really fast. otherwise SDXL turbo rules !
I just saw their website changed to Global Stardust, and I’m really upset... honestly heartbroken. My Yuanyong will never get to see my lover... 🥺
This shit blows. Even noob has a better understanding of language and tags. The worst checkpoint I've ever used and your asses thought you could scam people for this half baked garbage
Noob is superior anyway, because of vpred. They will release they v-pred who knows when.
Bro's never heard that base models aren't finished and made to be fine-tuned into a proper model
@Foxyz man have you seen the seething I've been getting for just saying "I'm just gonna put my experiences and ignore the drama, try it and make your own mind up on it", I'm gonna hazard a guess people currently some actually care about the model itself. But from the look of the actual ratings it's not the majority so.
I guess God forbid people actually use this comment area to help the people who have no clue and just want to understand the models strengths and weaknesses for on site generation, right?
I'm not gonna feed the outrage culture, or touch on the drama in itself, not my thing and if you're here for that opinion look at the countless other comments I'm just going to give a subjective take on the model;
I like it! It's quite good for a base model and plugging loras in have given me results I really really love. The ability to better prose with tag - story - tag is very welcome as that was my preferred method of prompting illu and now I no longer have to strictly adhere to tag NL; which as a primarily Google Imagen creator is a boon!
It has the downsides of any non finetune or merge of course but I've found it far more reliable than the vast majority of open source base models. :)
Take it as you will, that's been my experience it may or may not be yours but I say give it a go and make your own mind up on it.
Compatible with 0.1 character loras? Or that ship sailed.
@michaelatkin855 worked with the 0.1 loras I used but I assume milage may vary
ppl dislike the marketing of Illustrious and delve in rage for... a team that rooted the best ever open source base model ? they forget that nobody deserve such nice thing at all ! their selfish behaviour can't handle drawbacks of the one that works for them basically for free.
they prefer NoobAI with their shitty licence over the Illustrious one that is letting us do what we want.
training a 10 to 20M dataset at frcking 2048 pixel in both natural and tag is a daunting task ! i won't be the one complaining that they came back from the dead asking money ! as if it wasn't a money pit from the start... i'm just happy they still exist... unlike stability.AI
can u add the bralines tag to the model,makesure it is bralines/bra outlines,...........not bra visible through clothes which is bad style.
you cant do that without retagging thousands of images and training again
@derpmagician i am trying to train the lora,it is not easy,some SD1.5 ckpt can easyly made it without prompt(bralines),.i want to train useing these photos.
I think it's pretty cool that the lead dev on Illustrious is making posts explaining what they're currently working on: Illustrious Tech Blog - AI Research & Model Development
But, that aside, isn't it time you guys gave an apology for the misleading "Milestone = Model release" post and the subsequent backpaddling? It's very telling that the Stardust sponsorship has barely increased over the last few days, compared to how fast people donated to unlock v3.5 v-pred. I do believe people are willing to sponsor you and your work, but you have to regain trust and show that you are worth spending money on.
Do you think, they cant cheat with the sponsorship slider?:))))) Avid scammers, they have no interest to release 3.5 with open weight (500k bucks, jesus).
bro stop with the create only bullshit the model is already made and more already on the way I don't see the reason for doing it
You're not learning anything. Again I can't download the model. I don't fucking need online generation, I didn't buy my video card to generate images online.
Wait so now I can't even pay to download this? Not even on the website with stardust?
You couldn't do this with v1.0 and v1.1 versions from the beginning either, but after a while v1.0 and v1.1 became available for download. Hopefully they will do this for future versions as well
@MindInTheDigits I think I could use buzz though, right? Maybe I'm misremembering but I thought they had it in early access for purchase with buzz. I was OK with that, like fine everyone's gotta eat - but I want to make LoRAs off 2.0 using like OneTrainer and not via the Civit website and I'm not going to be able to do that I guess? Or hopefully they'll make it downloadable eventually I guess.
They might not release it unless the site’s stardust reaches its goal. Typically, when an older model file is made available for free, the next one becomes a paid release. That’s how it went with 1.0 being released for free and 1.1 initially being a paid version. However, the creator of 1.1 later made it free, saying the stardust had been updated. Now that 1.1 is free, you’d expect 2.0 to at least show up as a paid download, but it hasn’t. It’s hard to say how things will turn out with 2.0 since it isn’t even available as a paid option on TensorArt. There’s been no word about it, and the team isn’t communicating anything. Meanwhile, it seems the noobaixl team is planning to work on noob v2, v2 so it might be more promising to just wait for that instead
@dopml87641 Are the NoobAi developers really going to make a v2? Where did you get this information from? It would be really great if it's true
@MindInTheDigits This is discord information. Hello, esteemed AI developer community 🧑:
We are pleased to announce that LAXHAR LAB is officially launching the NoobAI v2 research and development initiative!
At the forefront of current technology, our multimodal pre-train model has made revolutionary breakthroughs in cross-modal inference, industrial-scale deployment, and multilingual generalization capabilities, laying the foundation for the next generation of general-purpose AI. If a raw multimodal pretrain model (meeting performance SOTA, open source license friendly, and multimodal dynamic fusion) is released by the open source community within 2025, we will fully commit to NoobAI v2 development!
We invite developers and researchers from the open source community to join us:
- Track frontline multimodal projects on platforms such as Hugging Face,
- share training methodologies (contrast learning, dynamic LoRA fine-tuning), and,
- participate in technical co-building and PR submissions via GitHub/GitLab.
The latest progress will be synchronized in real-time via Discord and the official Hugging Face repository, so stay tuned!
@dopml87641 Now if they need money they just need to say "give us some money" and I'll give them some money. I don't wanna play a gacha game. This really has made me wanna ditch Illustrious altogether which is a shame since i've made a bunch of LoRAs on it already but I suppose I have the training data sets so I can just remake them... Anyway! Thanks for the info - I'll keep an eye out for Noob v2!
@dopml87641 I think, that wont be on XL. OpenAi released they closed source solution, but its extreme resource hungry for local generating.
@aikage noobai have some shady licence tho... even if it is a great model. but they cannot enforce it anyway.
From testing out with it in online generation, its ok at best, but nowhere near NoobAIXL. Thoroughly disappointed that the model is not available for offline generation after the whole fiasco with 1.0 where the community strongly responded to 1.0 being unavailable to download.
Regardless this model doesn't bring anything new to the table anyways. So dissapointing.
please don't hide the sample prompt on the model display.
This should be illegal for models that are charging/were charging money. Essentially false advertising if no one can get results like theirs.
i'm glad OnomaAI is back, not surprised by the way they open source their model behind paywall at all. it's sad but guess what ? FINETUNING 2048x2048pixels IS EXPENSIVE.
training so many models is bound to become a HUGE money pit wheter we like it or not.
the advertisement sucked and they didn't delivered what we expected on time but now here we are !
they have a business plan that would allow them to sustain themself and pursue their research which makes me cautiously optimistic that they can succeed where SAI failled.
i'm happy to take any downvote for being based.
I know most people won't bother replying to your "based" statement here, but just so you know, most (possibly all) of OmonaAI's training hardware is subsidized by being in programs like the NVidia Inception program. False advertising is actually illegal in most countries, and you cannot technically by definition "open source" software behind a paywall and forbid distribution, which is their current stance (as of today).
@KingLord i don't know much on that fiasco and the law around and there may be more than meets the eyes. so please let me know if there is.
but as i see it from the paper, there is no legal issues. inception program is probably not free and open source program have been made behind paywall before.
welp 8 months later seeing how they fail at keeping a simple progress bar readable and transparent it's time to press forget n' move on. time to train QWEN instead...
Looks like crap compared to even 1 year old finetunes of other checkpoints.
Gens look far better than 0.1. Imagine if they actually opened the model up for download and let fine-tuners work their magic?
@ClownReplica too bad they went full greed. they cannot be trusted. when they saw their silly stardust meter fill up so fast they were like "but what if we made MORE money? 🤔". never trust them again.
@ClownReplica
imagination ... it neither buys us bread ... nor can it provide the characters for our stories,
we - unfortunatly - have to work within the fabrics of reality
and what's available for us to use is a non-downloadable, non-finetunable base
without any indiciation of it becoming frre and/or usable in the foreseeable future ...
so, that's what we can use ... so, that's what we compare it to
@ClownReplica Yeah, imagine what would happen if they actually do what they promised people they would do. Or what they advertised and took money from people to do. I'd love to imagine this company not lying to people and just taking their money to train and develop models and then NOT releasing them as open source to the people that are the foundation for their success - but sadly this is a pleasant fiction and nothing more.
1. On site generation only
2. Auction system
3. Review: not good enough
For someone who want to try, those three are a good combo to say "I skip this"
The first one really should be enough.
@Giangiotto well I can not justify only the first one, for those who had low VRAM (incld. myself) or no gpu at all, the first one is not significantly bad. But the trio make a too perfect combo
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
i still hope it gets released open source by whatever means ! at least V2.0. i want to see how a 2048x2048 model behave (in training and controlnet) even if it's bad / hard to use !
blender 3D wouldn't be open source to this day if there was no crowdfund back then. companies won't sponsor them if they don't get their cash back. so there we are.
they ccould have done MUCH beter, but nobody was going to do that kind of model anyways... no competition
@Le_Fourbe No competition:)))))))))))) What are you smoking? There is nothing else in anime field, but new porn models.
@Le_Fourbe True, and even if 3.0 comes, is it as good as 2.0?
For example, if something is broken in the newer version but still works in the older version, why wait for it right away? Same point. Blender 3.4 was fully functional as 4.0, and some of them are not actively making new versions for Addons, which could be a challenge in the future versions.
At this rate, you get your own mindcontrolled waifu, while they still gather enough buzz to release 3.5.
So much for the V3 models not being ready yet. V3 EPS and VPred are now available for image generation on their website.
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.
@kuraig in that case, why did you come here? continue using paid garbage that will never be open source.
what is NAI?
@simonrpayne9153 "neural network" (a merge of models that violates licenses), access to which is sold BY SUBSCRIPTION to madmen who don't like to think with their heads, or who have weak hardware
What is this corpo bootlicker doing in the open source community? If you like that corpo model so much, then why are you here searching for an open source model?
lmaaaao
Illustrious xl v2.0 has reached 100% of the fundraising, we expect its release in a few hours.
It released on hugging face and tensor art for download already
@DawnDenn It's the incorrect version, they seem to be keeping the real one and 'refined' it and sent out a different one
@Light7799 bruh thats so shitty...
Maybe they'll allow download on this page eventually? It's still weird and they're mostly radio silent about things. I have no clue.
Edit; now they have it downloadable
Tensor Art’s 2.0 version has been newly uploaded and is available for download.
This isn't the real one. This is the 'stable' version. They haven't, and I don't know if they will be releasing the 2.0 that we should be getting. On their official page on civit you'll find they have a 'stable' 2.0 version for download... if you compare the gens to the huggingface one, they're identical. I don't know what they're doing, honestly. Otherwise why wouldn't they just make this downloadable like they did with 1.0 and 1.1??? They made a whole separate checkpoint page and uploaded two of the same models on that page, but still no base 2.0
Edit; seems they finally made this one downloadable
GGUF Version: https://huggingface.co/btaskel/Illustrious-XL-v2.0-GGUF
Does this work in reForge?
@nawel34085394 I don't used Forge,but comfyui can run gguf
It says 3.0 is live but where is it? Its very unfindable.
It's only "live" on their website, using their own on-site generator.
Nothing of the user generated content looks anything like the examples. And you are hiding all the settings for some reason.
there is no point in looking at user generated content on civitai, 99% of them don't know how to do anything
99% of people who generate images don't know what they're doing...
real and true. looking at user examples to learn anything will just make you lose braincells. the fact people still use "bad hands" is extremely funny. i swear on god i once saw "the most beautiful woman ever to live" in a prompt. completely goon brained lmao.
@crombobular 太对了，这样的提示词真的没用
Illustrious v2.0 launched with so much fanfare. So much so that I never knew it existed. Lmao
Nobody use it. Finetunes are miles better.
Illustrious launches are the most confusing things ever. This page is listed as MARCH and it's nearly JUNE and I only just noticed this exists. And I'm on Civitai every dang day.
Bruh
Well, I'm guessing we aren't getting releases for new model versions after V2.0, seeing as how V3.0 was supposed to release in April and V3.5 in May. Guess they were talking about they're own on-site generator.
Nothing but lies and empty promises from these guys.
yep its completely behind a paywall now, im sticking with 0.1
Diffussy v0.1 is only good for its architecture and functionality. For actual generations, use fine-tunes instead.
Scummy business move resulted in less than 1% of contribution from the community to the next 3.0 release. Can't say I'm surprised. Noone could have ever imagined this outcome 6 months ago with the revered Illustrious name. What a quintessential stable diffusion moment indeed.
dead loop. setting up a name, a website, a business entity, and try to grab users to take their money. eventually you see all of them going into the same route. a dead loop that doesn't work. individual or corporate model maker. all of them.
government/central bank control the coinage right, hence the law of anti counterfeit measure, why are you trying to get money from personnel that do not hold the right of coinage? go after government. you will see the national annual budget meeting is the real fight, and you will also see the fight for money is ultimately a dead loop as well.
budget is the government licence/permit essentially, you get to spend it in the following year, ultimately you need to produce something, that is your responsibility, things you are suppose to do.
you can't eat money, and money won't do anything by itself, you guys are getting the ass kick from CN cause CN actually made something with their budget while the rest of you were stuck inside this dead loop. the arm making drug making industry, and their cute buddy bankers, are the most paper rich group of people in the entire world, but you are getting the worst medicare and arm supply, it is the same logic in diffusion model making, people who make the worst model are those money grabber.
it is the money diabetes. beyond reason and cure.
i agree
Although it can be understood this way, its open-source protocol seems to be more beneficial to creators than NOOBAI's
Never expected to read something to this degree of lucidity at such a random place. Being able to see things so clearly is quite the burden sometimes, so I hope you are faring well, brothar! More power to you.
Are you planning to release version 3.5 to open access? On your website, it already says 329,220 / 300,000 Stardust raised
that is not for the release of the model of illustrious v3.5 vpred. that stardust is proly for something like test license or something. on their website there is another stardust donation and that was in millions that’s where the target have to be reach for model to be released publicly. Honestly the way they didn’t mention anything about where to donate for what is very misleading and could have made others donate stardust on wrong thing which has happened at its over 300K+ stardust on illustrious XL 3.5 vpred model.
Honestly we should all cancel pony diffusion and Illustrious XL for going down the money path and selling models and their own money goals in name of open ai for all. this is no different than paying for chatgpt and google gemini.
I hope someday it will get released.
KookiePum I would like more versions of illustrious, but I think it's fair to ask for donations. It costs money to train the models, after all.
AI的天才营销的蠢材，我只能这么评价，完全可以通过提前释出新模型来赚取口碑和金钱，偏偏选了个“呃呃，那个进度条不是release目标哦”，长脑子了吗？是人类吗？
I have dowloaded 3.5 from piratebay. Its size is 35 megabyte. Is it genuine, or virus?
100% virus bro 2.0 is 6.5 gigs xD
Oh boy you are cooked
Do not redeem
Its genuine. I can run with my first videocard(geforce 2 mx 400 64mbyte) easily. Its better, than anything else.
What's the thinking behind this brother?
bro went silent, they got em
Progress bar is not moving at all. They will never reach the "threshold" to release 3.5. Maybe in the future, when SDXL will become obsolete. Naiv3 will go open source sooner than this lol.
wai-shuffle-noob v-pred-02, vs ,
WAI-NSFW-illustrious-SDXL-v14,
which do you prefer more?
I have the images here, but I’m not sure how to send them.
Anyway, on July 5, 2025, the points showed 1,567,625.
Today, on September 20, 2025, the points show 1,568,791.
Your announcement states: "Every image you generate contributes to the Global Stardust Progress Meter, not just direct sponsorships."
This either means you’re secretly manipulating the progress bar to keep it almost unchanged, or it means nobody is using your online services.
If it’s the latter, your plan is pretty much a failure.
lol😘
Hi!
It's been 5 months and you ONOMA still being a massive pusi.
Just to remind you.🥰
This model is just an ad for something no one will buy into.
people download models to use them locally, why would we use an onsite generator by a company that screws people over then goes silent about it?
Illustrious XL v3.5 vpred has reached the stardust goal with 329,390/300,000, do you plan to upload it, or do you want to wait for previous versions to also reach their goal to be able to upload them? Because honestly I don't see it likely that those models will reach the established goal.
Pretty much Illustrious model is DOA.
that donation goal is not for model release, but that donation goal is for their high tier supporters A.K.A they have already given the access of Illustrious XL v3.5 vpred to thier top donaters only. So, according to what they are claiming that for public release or for people like us we need to complete a totally different donation goal which is in millions (Literally millions of real money) for releasing it for public use. And I am pretty sure that their money hungry butts will release a badly epoch of model and will keep the best epoch behind pay wall, like they did with Illustrious 2 and Illustrious 2-pro model.
Pretty scummy business tbh.
Disclaimer: Not a defender of ONOMA's stardust practices. I fully know that this will be downvoted, but I'm just stating what their site says.
"Illustrious XL v3.5 vpred" being 320k+ out of 300k, does not equal to being an open-sourced released model.
This goes under "Each model has its own Individual Stardust Meter, reflecting its specific funding progress. When an upcoming model’s Individual Stardust Meter reaches 100%, we will provide academic testing access to those who contributed."
However, "Illustrious XL v3.0", which is under the Released category, is at 176k/300k (which means it's off by 124k = $1240~). This model category goes under "When an open-sourced model’s meter reaches 100%, we will release exclusive aesthetic-tuned checkpoints for testing to those who contributed."
1000 stardust = $10
At this point, it would be better for OpenAI to release an open-source version of Sora, or for Qwen to release a better image generation model, than for them to release Illustrious 3.0.
your omnoai team is all sure about still keeping illustrious 3.0 and 3.5 behind paywall and donations goal? because what i am seeing is that your website is barely getting 50 monthly visits of traffic. You guys are only going in lose for a year now. Better to go open source with what you have than insisting over money goals.
Went to the site lol It's Just a wrapper for Gradio lol. I made one in like a day for free ( with alot of gpt help )lol. But I don't see anything of value. I'm not signing up to generate images using a Illustrious Checkpoint that ( may or may not ) be just a merge of other stuff lol
DPM++ 2S Ancestral CFG++ what the sampler?O_O
All images I generate with this checkpoint look blurry and distorted, and they do not resemble the example outputs at all.
For reference, I’m using:
- Sampler: Euler A
- Scheduler: Simple
- CFG: 5
- Steps: 30
There also still isn’t a comprehensive prompting guide, which makes the model much harder to use effectively for power users trying to get reliable results.
ONOMAAI
What a scandalous business:
1. Deliberately released bad epoch of Illustrious XL v2 as open source and gate keeping good epoch as Illustrious XL V2 Pro behind way walls.
2. Not releasing of Illustrious XL v3.5 after reaching the donations goal.
3. Stealing Anima Preview models and putting it behind paywall on your website as if you are the one you created it. But no, it is creation of Circlestone-labs and ComfyUI.
You, money hungry Koreans are deserving of the coming downfall of Illustrious XL as whole genre.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.