Asmodeus - the prince of demons; representing Lust...
This checkpoint is capable to produce nearly everything you can imagine... It's completely uncensored. The main focus here was creating photorealistic pictures but it can create nearly every style.
Asmodeus is NOT created to produce just porn! (you can; but don't share it here; nude art is ok though)
If you like it, feel free to give me some feedback. Also if you want to support me you can do this here.
https://ko-fi.com/afroman4peace
Version 3.0:
Well, this is the final version of this checkpoint. Asmodeus is now merges with DEMONCORE "Research" https://tensor.art/models/724032586164281309 and Midgard 1.3 https://civarchive.com/models/463847/midgard-thl-hybrid. (I highly recommend to check it out too.) Thanks to freek22.
This model combines creativity with photorealism. This model is made for things like creating nude art. You can of course use it as normal general checkpoint but be careful with it.
Don't let your kids play with this thing! NSFW images can be shown without prompting it.
This model works on a high range of cfg and steps. ( I will add more details soon).
Use about 10-16 steps.. (DPM++ SDE)
I recommend using DPM++ SDE (Karras or normal) at 3-4 cfg
Don't use DPM 2M or Heun!
A workflow will be added soon.
Known bugs:
This checkpoint can produce skimpy or even nude images by accident. If this happens (mostly when you describe the skin our breast size). Add nude, tits, NSFW.. to the negative prompt to prevent this. Be careful with the word "girl"
Do not attempt to produce naked man; seriosly don't! It will not produce male genitals
Version 2.2:
This Version is a very complex merge between several my Hephaistos (many of them are not public) and Asmodeus 1.7. This version behaves a bit better and the textures eyes and faces are sharper.
Feel free to try all versions and tell me which one do you prefer.
Why version it began with V1.7?
This checkpoint is based on my XL6 Hephaistos model V3.30 (which isn't avaible yet). The goal was to create a much more capable model regarding realism. I did this by adding several Loras and my own training data. By adding NSFW Loras I improved the anatomy of humans a lot (regarding faces and other things). The first result was terrible at first; just body parts and "human blobs". So I added some parts of my other model I am working on... this resulted in good pictures. But mostly naked woman and no art. So I remerged it with Hephaistos 2.1 and freek22s FenrisXL. (I highly recommend this checkpoint he created on base of my Hephaistos model). This stabilized it and now it works - with some flaws.. but it works.
Disclaimer and warning:
This checkpoint is fully uncensored. Only your imagination is the limit. Beware of this.. It's not intented to create just porn but it can be very horny. I would advice to consider this and maybe add NSFW in negative prompts... Don't use this to harm others... Please use this responsible! You are fully responsible for your actions!
And most important:
Don't use this to create CP! If I see such pictures here in the gallery I will flag it and you might get banned from here..
Useful tricks:
Don't use the refiner! It seems to causes only bad results.
On standard this model trys to create photos. So there is no prompt needed for this but it can enhance the results. play with different cameras and settings.
You (almost) don't need negative prompts. I mostly use the prompt "boring". To handle the skimpy behaviour of this model try to use words like tits, nude, NSFW in the negative. Sometimes especially for creatures like cats or mythical creatures add woman to the negative prompt.
(V2.2 and V1.7) I mostly use the 2M SDE Karras for my prompts at around 40-100+ steps with a cfg from 7.5-10. In my tests it showed that more steps (up to 300 steps) can increase the quality. https://civarchive.com/posts/541429
Feel free to experiment here. I haven't tested it very long.. only about a week.
Asmodeus can handle resolutions up to 1200x1200 very well. You can go above but it can increase your failure rate
To upscale your results you can just use img to img. use around 0.2-0.3 denoise. (this can vary depending on the prompt) feel free to experiment with it.
This is a workflow for Comfy UI https://civarchive.com/articles/1978/workflow-for-xl-13-asmodeus
Have fun
Credits:
Midgard by freek22
Finenude V.02 by finenude
Hella Babes, a SDXL LoRA for better nudity by vfx_ai
Film Like XL by getjinxed
Fenris XL by freek22
Description
Don't let your kids play with this thing! NSFW images can be shown without prompting it.
This model works on a high range of cfg and steps. ( I will add more details soon).
Use about 10-16 steps.. (DPM++ SDE)
I recommend using DPM++ SDE (Karras or normal) at 3-4 cfg
Don't use DPM 2M or Heun!
A workflow will be added soon.
FAQ
Comments (13)
Go ahead and look again the checkpoints you did last autumn. They're all far superior in quality than any of these accelerated models you've released this year. These new ones all have the same flaws. Details and compositions are great, but they're all ruined with muddy output that lacks clarity and sharpness.
Thank you very much for your honest feedback. Yes I recently did exactly that. This is why I am working on this checkpoint again. Tell me if this got the same issues. I may create a normal not accelerated model soon.
Some of the new checkpoint need specific samplers to work correctly.. This is caused by some of my training. 2M sampler and Heun doesn't work for example. Also PLMS and Restart doesn't work. The results describe that what you told above.
If you don't mind.. can you send me some bad examples? This would help me to improve my checkpoints. Or what sampler /settings have you used?
@friggensarrr What also can help me.. What is your favourite checkpoint I made last year without this issue? I think this could be also beneficial for future checkpoints
@Afroman4peace I deleted most of bad images already. I don't it's worth to post them. I tried all the usual dpmpp samplers with normal, karras, exponential, sgm_uniform. Tried cfg 1-5. This asmodeus 3.0 works for me the best with euler_ancestral and sgm_uniform. It was the first one I tried, because I've tested many sdxl and pony lightning checkpoints and almost all of them work the best with that combo. There are couple that works better with Euler sometimes. Actually, asmodeus 3.0 isn't so bad. I should change my review to positive, because I think it's worth to test it again with different kind of arts.
@Afroman4peace Hephaistos v3.3 is my favorite, or v2.1. I'm not sure which one is better. Both generate lovely steampunk, cyberpunk and robotic creature images, or anything futuristic really well. I don't think I had any problems with them. In general, I would say all your older checkpoints are really good.
Lightning lora's are not perfect though. They have facial detail problem. They have that one face that appears all the time. Better use character lora's or established characters, or you'll see a lot young Jeri Ryan lookalike blonde with it. Lightning lora's also generate sometimes pimples or strange blemishes, or scars on the face under the eyes for no apparent reason. Removing any facial detail tags like high detailed face or skin from the prompt helps a bit, but not always enough to my liking.
@Afroman4peace Maybe it's not a good idea to merge different accelerator techs together in checkpoint. Maybe there lies the problem. There are some horrible pony checkpoints with turbo hyper lcm merges that can't generate a cat.
@Afroman4peace Thanks you very much for huge buzz tip.
I completely rebuilt the visuals.. I used a old but a bit further developed Hephaistos and parts of the 1.6 Colossus version. Then I stripped a new not uploaded Colossus version from the visuals and exchanged it.. The error was maybe due a faulty harddrive back then.. I will upload some images in my gallery..
@friggensarrr here are some first results. https://civitai.com/posts/2982497 So I basically used blockmerging here. Sadly I need to retrain some parts of the visuals.. which will take a lot of time.. but I have the datasets ready
IMHO. I have hyper model can do good more only prompt controll not well.... less quality for me...
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.