✨ Tired of typing prompts + want viral-ready art?
Pair Lazy Embeddings 🛋️⚡ (zero-typing magic for all models ) with TrendCraft 🪄🚀 (30-day trend-trained detailer ) for effortless masterpieces. Check them out: Lazy Embeddings + Trendcraft (Pony XL , Illustrious , Noob Vpred, or SDXL base trained version)
👉 Join me on our discord server (https://discord.gg/tWrKk3jcXx)! Share art, ask for help, or just chat. We have CivChan bot plus games there!
👉 Support future updates: Buy me a coffee or tip buzz! Every bit helps. Thank you for fueling the art! 🚀
🚨 JANKU 🌟
✨ Key Features:
🔥 LoRA-Free Workflow: Generate high-quality art without stacking detailers (unless you’re adding niche styles/concepts).
🔞
𝐍𝐒𝐅𝐖Mastery:Better anatomy (yes, even in complicated poses) + improved male framing.
𝐍𝐒𝐅𝐖: if you want more control, put lazy𝐍𝐒𝐅𝐖in negative!
Large character list support: Character list shared in our discord server.
No more stability issues —generate confidently at 1536x1536 without up-scaling. My personal resolution to use is 1024 x 1536.
🎨 Crisp Details: Enhanced backgrounds, outfits, and character features.
👀 Prompt Responsiveness: Gets your vision right the first time!
🔧 VAE Baked In: No extra steps—just generate.
Read version details for the version you are using for more information!
✨Improvements from RouWei
Easy and versatile prompting with great aesthetic, anatomy, and stability
Vibrant colors, smooth gradients, and full brightness range
Over 35k artist styles pure training (no merged weights), and customizable options. The list in .txt can be found on our discord server.
Note: you must use
by artistnameto trigger it!
Eliminates watermarks, tag/character bleed, and strange artifacts for better coherence
Improved prompt following, stable styles, and accurate artist representation
🛠️ Recommended Settings
Steps: 25-30 (I use 30 normally)
CFG: 3-5 (I use 5 normally)
Sampler: Euler / Euler A (my go-to)
Scheduler: Normal or Simple (my go-to) or SGM UNIFORM
Model Recommended Resolution：768X1344 832X1216 768X1280 704X1408 1024X1536 (my go-to, no need for up-scaling!)Standard Illustrious & SDXL resolutions also supported
Prompting: Use my Lazy Embeddings for positive and negative and even 𝐍𝐒𝐅𝐖embeddings to prevent or add 🙈. Use lazywet to remove wet and shiny skin. No more guessworks for prompting! If you don't want to use embeddings, use your normal NoobAI/Illustrious quality triggers. There is artist, danbooru, and rating tag support.
Positive promptlazypos {character tags} {detail tags related to character} {background tags}lazypos, 1girl, makima \(chainsaw man\), yandere stare, smile, dress shirt, grey jacket, red hair, sidelocks, black necktie, looking at viewer, jacket, long sleeves, upper body, suit, necktie, shirt, simple background, medium hair, collared shirt, ringed eyes, shaded face, braided ponytail, head tilt, braid, white background, formal, yellow eyes, holding own arm, wing collar, closed mouth, solo
Negative promptlazyneg
Pro Tips
Using my Lazy Embeddings for onsite gen doesn't cost you extra buzz, if anything it helps you save buzz since you're not generating bad images!
Check my images’ metadata for exact prompts/settings. Import them into ComfyUI to reverse-engineer my workflow!
🩺 Fixes & Feedback
Eyes still wonky? Use eye detailer, but most of the time you don't need unless the subject is further away.
Found a bug? Report it—I’ll fix it in next version. I only generate with this model, so I’ll keep improving it. I won't publish something I won't use myself.
❤️ Why I Made This
Tired of juggling 10 LoRAs for 𝐍𝐒𝐅𝐖 art? Me too. I want a 𝐍𝐒𝐅𝐖 checkpoint with more knowledge of characters and other common booru tags. This checkpoint is my personal playground —designed for creators who want bold, detailed, no-BS generations.
Future updates? Always. Let’s push boundaries together.
Please share your feedback to improve the checkpoint together! ❤️
JANKU Checkpoint License & Terms of Service
Comply with the NoobAI-XL 1.0 License
User Responsibility :
You are fully liable for all outputs. Use prompts to filter content.
Restrictions :
No redistribution of modified versions without adhering to open-source terms.
No use for unlawful or harmful purposes.
Failure to comply voids usage rights.
Description
v3.0 released!
Merged v2.1 with RouWei base. Lots of fixes added to adjust brightness and contrast without prompting! RouWei base has updated knowledge cutoff at Dec 20th 2024.
Merged in some blocks from Janku v2.0 and v1.0 to bring back some more anime style.
Better shade lightning compared to previous.
Brighter colors, less dark.
I used new merging strategy that I learned when I worked on Jankun vpred version of the model. Happy to bring this to Janku!
✨Improvements from RouWei
Easy and versatile prompting with great aesthetic, anatomy, and stability
Vibrant colors, smooth gradients, and full brightness range
Over 35k artist styles, pure training (no merged weights), and customizable options.
Note: you must use
by artistnameto trigger it!
Eliminates watermarks, tag/character bleed, and strange artifacts for better coherence
Improved prompt following, stable styles, and accurate artist representation
👉 Join me on our discord server (https://discord.gg/tWrKk3jcXx)! Share art, ask for help, or just chat. We have CivChan bot plus games there!
FAQ
Comments (53)
v3.0 released with RouWei base merged successfully with brightness and contrast fixed. Along with previous version merges to bring back more anime style. Please read the version details for all the changes. I have moved all the version changes for each version there.
This person is associating minors with adult material.
Really? Is there any evidence for this?
@Oskarjasz Yes, they talked about exploiting tags to get "slut" themed images into PG so kids can see them, highlighted by images with "SLUT" in big bold text which is adult themes relating to sleeping around. Plenty of evidence, avoid their discord at all costs.
v3.0 onsite gen next week 🤞. I will bid for it for all the onsite gen people due to large amount of early access support ❤️ It would be in early access for a few days during it, but I will see if support can end it early. v2.1 will be for sure as an amazing supporter bid for it already! Thank you again ❤️
About artist triggers.
It seems that it's the same as for the other illustrious/noob models. You can use ''by artistname'', ''artistname'',''artist:artistname'' and it will work... but the results will be different. Not even better or worse, just like a different seed was taken.
That's not bad, just slightly annoying, because you have to try all these different prompting schemes to get something you are fully satisfied with.
As for your chekpoint - seems ok, i will test it more in the upcoming days/weeks.
Thank you for your work.
Thank you for your support! You need to use the right tag based on what the model was trained on. Any words you put into the prompt will change the output even if the model don't know the word or trigger. If you have v3 use the list provided if you have using v2.1 use the booru artist tags.
v3.0 and v2.1 both have onsite gen added. Thanks again for the bidder that helped with v2.1 ❤️ Make sure it is not draft mode. Use 25-30 STEPS and 3-5 CFG with Euler A! Looking forward to what everybody creates!
I am really surprised!
I have tried many checkpoints for a while, and I want to say that this one is the best without any doubt. I love that it doesn't have those annoying quality tags. And the quality and color of the images is awesome. If you are looking for quality, look no further, you have come to the promised land when it comes to checkpoints, try it I promise you won't regret it. It is the best checkpoint you will ever use.
@Separate_Ways Thank you for your kind words, it means a lot to me especially coming from another creator like yourself. I will continue to improve the model 😍Always happy to take feedback. If I don't see it please free to DM me on discord. Server is in my profile 👍
@janxd Recognition to those who deserve it and you deserve it!
👉 Join me on our discord server! We have CivChan there plus discord games!❤️
Astonishing! I Tried 7 different checkpoint for last 3 days and this one give me more the others combined!!!
Hi, I can't get the artist list txt file, the link seems broken here, can you fix it please (if it's really broken) ?
@Brewce I just loaded it's not broken. I can upload it through discord if that works better for you. Find the link above or in profile!
When using this Checkpoint, I always have trouble getting the gray eyes right on the characters.
If character is far away, I suggest you use a/detailer with eye details prompting.
Thank @FireDextro for all your support! Working on the next release! Check out the discord if you want a sneak peek 😉
That looks like conditioning to and from so it is related to the tool you use, not the checkpoint. I can do this conditioning in comfyui with custom nodes, no issues.
https://github.com/asagi4/comfyui-prompt-control is what I use to do that
Absolutely cracked model, fantastic job. Can vouch for the "improved male framing".
Seems to have lost a little bit of NoobAI's knowledge of niche e621 tags, but this is to be expected with merges, and it makes it up it with much more coherent backgrounds and the ability to do 1024x1536 natively.
Works very well with artist tags that I tested on models that didn't work before. Probably my only issue is hands sometimes came out a little worse.
Friendly reminder, your model has lost almost all the eyes and facial features of anime characters.
Please share your prompts and pictures so I can help you. Please check the pictures below as well for more guidance.
I mean your modle output with almost the same face, you can try Obsession (Illustrious-XL) for compare. The charactor I tried include kurokawa_akane, faruzan_(genshin_impact), power_(chainsaw_man), souryuu_asuka_langley, seele_vollerei
I generate lots of Power before with no issue. What else is in your prompt? Try using 3d, realistic in negative.
You can refer to below link, I generate using Obsession Illustrious and with my own lora
Showing me an output of another model while using lora can't allow me to help you. I just tried kurokawa_akane, faruzan_(genshin_impact), power_(chainsaw_man), souryuu_asuka_langley, seele_vollerei and got correct outputs with those tags and my embeddings nothing else. Try using my lazy embedding: lazympos in positive and lazyneg in negative.
Are you saying on the pictures generated by people below including me have the same face?
Could not recognize penis's foreskin,looking for update.QAQ
Amazing! Now I use only this Checkpoint. It helps to create very detailed images. Thank you! ^^
Thank you for using the checkpoint 😍
The images you generated with that model look great, but when I try to remix with civit generator I get far worse results, far worse style compared to your results. Any idea why? I also tried using Comfy and it still wasn't same style as your generations
Onsite generator make sure it is the same settings as the image, it should be close with remix as that is image to image. For ComfyUI try importing my images the workflow is included. You can never generate the same exact image unless the computer is the exact same, for example I got a new GPU and I can't generate the exact same images as before but it should be close (applies for everything). Onsite gen doesn't use the same prompt parsing setup.
@janxd interesting, I'll try importing your workflow and check again, thanks! Great checkpoint either way 😁
Do you have any plans to update V-pred?:D
@miss72913_m1 It's on my list to try on merge Rouwei vpred into it along with lighting fix. Biggest issue with vpred it sucks for onsite gen and most people don't know they need to update their local to use it correctly or use other settings to adjust the output. Since you asked I will move it a little higher on the list 😉
@janxd I'd love to see a v-pred version too
This easily became one of my favorites!
Thanks for checking it out! ❤️
Pretty model, but it seems way too weighted to producing forehead-exposed hair styles, even if I specifically ask for some sort of bangs.
Really good to be honest. My most frequently use model is Wai-nsfw-IL so take that for example.
It is more like the combination of wai v9's style creativity with wai v11's prompt accuracy and image stability.
But there also a strange weak point, personal feeling the accuracy of PENIS is little worse than wai, such as double or no balls with balls prompt.
One model sloved, now this one is in my favorates. Looking forward to next update, keep this great work.
The checkpoint is a masterpiece, really great work, thank you very much.
A few comments:
Since I used WAI-NSFW-illustrious-SDXL a lot, I want to compare the two.
Anime Screencap looks better with WAI for my taste (tested with: Fine Anime Screencap XL | Anime Screencap Style LoRa Illustrious and PonyXL)
JANKU makes stronger lighting and shading even without being prompted. That's also the reason why if you put "3d" with a high weighting in the negative promt it still goes a bit towards 3d in JANKU.
I like the colours a bit better in WAI.
I think JANKU could be a bit better than WAI at generating 3d and semi 3d artwork, especially with lighting and shading. The checkpoint can do that very well. But I haven't compared that much yet. Also it is i think generally better at artworks while Wai is better at nsfw.
In any case, a very impressive checkpoint and better than all the others I have tried besides the two mentioned.
Especially with TrendCraft and Lazy Embeddings you can really get effortless masterpieces as he says in the description. It is really amazing how easy you can get such good looking results.
Works extremely well with all sorts of LoRAs, simply amazing.
The images produced by this model are very clean and neat, well done
This might be one of my new favorites, great work! The anatomy is phenomenal and the adherence to my prompts is fantastic.
Very good 💯
Details
Files
JANKUTrainedChenkinNoobai_v30.safetensors
Mirrors
JANKUV4NSFWTrainedNoobaiEPS_v30.safetensors
JANKUV40TrainedNoobaiEPS_v30.safetensors
JANouwei.safetensors
JANKUV30NoobaiEPSRouwei_v30.safetensors
JANKUTrainedNoobaiRouwei_v30.safetensors
JANKUV30NoobaiEPSRouwei_v30.safetensors
JANKUV30NoobaiEPSRouwei_v30.safetensors
JANKUV40NSFWTrainedNoobai_v30.safetensors
Available On (3 platforms)
Same model published on other platforms. May have additional downloads or version variants.