Simile = comparison, likeness, parallel, similar, resembling
Note: Preview images are generated without the use of LORAs
NoobAI ver. [v3.0, v4.0, v5.0, v6.0]
⚠️ATTENTION⚠️
Check 'about this version' tab for more details..
Also, here is the list for the artist tags. There are two files in the zip file, one if the full artist list and the other is the list of artist that I've personally tested to be working and most of them are also from gay artists..
About
To be honest I didn't expect to update this quickly but I can't resist NoobAI's updated datasets. Artist tags also performs much better with less dependent on emphasis to achieve a desired effect. Overall it is an improvement of the already great model, Illustrous XL.
Tips for artist styles / characters
Similar to ILXL, just add in the artist or character tags and that's it. If you're not sure what to write for the tags, just search it on danbooru or here. See sample images for examples on how to prompt the tags..
My Settings
Sampling Method: - Euler a (less detailed and smoother textures)
- DPM++ 2M (more detailed lines and textures)
Sampling Steps: 20+
Upscaler: 4x_NMKD-Siax_200k / Remacri
CFG Scale: 3 - 5
Clip Skip: 2
VAE : Already built-inPositive Prompt:
very awa, masterpiece, best quality, absurdres, perfect quality, detailed background, intricate detailsNegative Prompt:
lowres, worst quality, bad quality, bad anatomy, signature, username, logo, watermark, artist name, bad handsIllustrious ver. [v2.0]
About
Personally I still prefer the default style of the Pony version but the prompt adherence and style flexibility of the Illustrious XL is just too good to pass up. Prompting and tagging is also a lot easier than the pony version from what I've tested. I tried to replicate the look of the Pony version but there's just too much artist styles for me to do anything without messing with the overall quality, oh well..
Tips for artist styles / characters
Illustrious is a fine-tuned model trained on Danbooru datasets, so there are a lot of data on various artist's styles and characters. So you can prompt a specific style or characters without the need of loras as long as they have substantial amount of dataset on Danbooru and are not too recent. See sample images for examples on how to prompt styles..
My Settings
Sampling Method: Euler a, DPM++ 2M / DPM++ 3M SDE
Sampling Steps: 25+
Upscaler: 4x_NMKD-Siax_200k / Remacri
CFG Scale: 3 - 6
Clip Skip: 2
VAE : Already built-inPositive Prompt:
masterpiece, best quality, absurdres, newest, perfect quality, best quality, absolutely eye-catching, detailed background, intricate detailsNegative Prompt:
lowres, worst quality, bad quality, bad anatomy, sketch, jpeg artifacts, signature, watermark, artist name, oldPony ver. [v1.0]
About
Just like with my HomoSimile model, I'm trying to get that specific look that's between 'anime' and 'western art'. I went a bit easy with the LORAs to ensure better compatibility with other styles. So if you want to use a specific artist's styles it wont diminish the results.
My Settings
Sampling Method: DPM++ 2M / DPM++ 3M SDE
Sampling Steps: 30+
Upscaler: 4x_NMKD-Siax_200k / Remacri
CFG Scale: 4 - 6
Clip Skip: 2
VAE : Already built-inPositive Prompt:
score_9, score_8_up, score_7_up, highly detailed, intricate details, detailed backgroundNegative Prompt:
score_6, score_5, score_4, score_3, score_2, score_1, low quality, deformed, bad hands, watermark⚠️DISCLAIMER⚠️
Please do NOT generate illegal content with this model. You are solely responsible for any unethical use of this model(s)
Model License
Description
FAQ
Comments (20)
Your Grace, are you planning to put the Illustris version on your TensorArt page? :-) Again, thank you very much for these outstanding checkpoint models
Lmao, sure. But it might take a while though, since uploading models to TA is 10x slower than Civitai for some reason..
@Kairen92 I guess because the servers are located in China...
@GloberTrotzer Ahh, I see. Well its at 15% now so hopefully no interruptions lol..
Thank you for the IL version! Looks like this checkpoint has knowledge of foreskin so this one will be my default for now! By the way, how do you prompt the for the style by artist kuromine hiraya/yuatari? They go by different names and I can't get their style to show up in Hoseki's checkpoint without LoRA's. (Downloading this one right now so haven't tested it yet)
I tried 'kurominehiraya', 'yuatari' and their japanese name, doesn't seem to work as well. On danbooru, they only have like 39 images so that's probably why. From what I've tested, artist with less than 50 images on danbooru doesn't seem to work that well. No choice but to use Lora unfortunately..
@Kairen92 Thanks :) I think I'll just do without style LoRA's for now. This checkpoint works great for generating what I want.
@Kairen92 This is late but I figured it out. The correct syntax is "by [artist name here]" so "by kuromine hiraya" worked for me. Specifically changing the weight of the word to 1.5. Unfortunately it's based on his old stuff which I don't think looks as good as his more recent works so... probably need to use a lora to achieve the look I want.
@cwhk That's good to know. And yeah, Illustrious is trained on stuff before 2023 so anything newer than that are not in the training dataset. But I am planning on making a quick update on this one that also includes 2024 stuff, but it might take a while though since I'm working on something else..
Do you have activation tags for Attack on Titan characters? And, if yes, which ones? Thanks a lot!
No activation tags, just write their full name and you're good, and maybe adding 'shingeki no kyojin' tag in your prompts too. But it only works on popular characters though like the main characters. So side characters with only a few of their images online probably won't work as well..
@Kairen92 Thanks. So I guess many tags are based on the pre-existing tags of Illustrious XL? (like styles etc...), or did you insert art styles (Suyohara, Phaustok...)? If yes, could you please write down a list or a pseudo-list for the users so we can use the checkpoint to its fullest? Thanks again for the great work!
@GloberTrotzer Illustrious was trained on booru tags, so if you want to prompt certain characters or styles, you can check danbooru to see if the tag exists. I use this extension to get the exact tags in A1111 so I don't need to leave my webui: https://github.com/DominikDoom/a1111-sd-webui-tagcomplete.
It's a bit outdated though and doesn't have newer tags. You can download a .csv for new tags but those might be TOO recent (illustrious doesn't have the latest tags)
I'd love to see an update on HomoSimile Pony. Illustrious is just not very flexible or particularly knowledgeable, though this is one of the best Illustrious models I've tried. Illustrious has more promise than it's able to deliver right now though, I think, where Pony is a lot more developed, so it's easier to get better results. In any case, I'm always glad to see more from you.
That's weird, it's usually the other way around. If anything IL should be more flexible and knowledgeable than Pony since its trained on more datasets and not just the furry stuff. I will say though, Pony is trained on a mix of 2d, illustration and realistic images while IL is mostly just 2d, so for other stuff it might seem restrictive but for 2d images I think it works the best (for now).
I'm not quitting on Pony yet though, it's still an amazing model. But for HomoSimile, I think I'll stick with Illustrious for now, until something better comes along lol..
@Kairen92 It should be, given its specific focus in training, but it isn't! It's definitely not better for 2D images, which are really the only thing I'm interested in, so I've focused on that more than 3D in every type of checkpoint. Most of what Illustrious puts out are garish, sloppy, overly undetailed, or plainly ignorant of tags. Maybe I just haven't found the magical Illustrious checkpoint that will do what I require from it, but so far I don't find it especially impressive. Maybe I'm expected to want to imitate other artists, which doesn't interest me. As your checkpoints are usually consistent and have a specific focus, I can still see appreciably better performance on your Pony models, and Pony models in general.
Testing this right now and is is amazing and really easy to use, I love it. Ooc do you plan to make one based on noobai in the future? The character and tags recognition are really nice on the last models
Oh man don't even get me started on that. I was just finished working on this one when I learned that NoobAI has the latest datasets.. 😩😩😩
For now I'm working on another model but after that I'll probably do a quick update on this model..
@Kairen92 it's absolutely amazing anyway. Noob and illustrious are already really gay friendly but this is the first male oriented model I test of them and it's a bliss what can do and how easy is to. And without random girls popping 10/10