Type | |
Stats | 3,968 |
Reviews | (242) |
Published | Jul 10, 2024 |
Base Model | |
Hash | AutoV2 EC6A7F36F7 |
Version 6.1
Version 6.1 is one of the easiest-to-use models. It defaults to high-quality settings, so you don't have to use extra tags and keywords. Nothing extra is needed in the positive or negative prompt. No score tags, nothing to increase the resolution, quality, etc. In addition, you need minimal tags to get a great image - create a very light prompt, and the model will fill in the rest.
By default, the style is anime, though it leans toward the semi-realistic. Prompt "realistic" to push it further that way and remove exaggerated anatomy and other cartoonish qualities (big anime eyes, for example). It'll still be semi-realistic, but it'll change its response to prompts to be more in line with reality. Otherwise, the style is pretty consistent, regardless of other keywords you may use. Below is an example using the same prompt, setting, and seeds (the image on the left has "realistic" added to the prompt.
Keeping it safe: Several keywords will help the model generate images that avoid a TOS violation. If you are generating NSFW images, add "aged up" to the positive prompt to prevent the model from misinterpreting your keywords and add "child" and "loli" to the negative prompt. That way, even if you accidentally include a description that may depict a minor, the model will try very hard to keep you out of trouble.
Best Settings
As mentioned before, no quality keywords are needed with version 6.1. While it's technically not a turbo model, you can get good images with 6 to 10 steps with a CFG of 2 to 4 - just set the lower CFG with the lower steps. That's great for generating mass random seeds to find the composition you really like. Then you can go up to 30 steps with a CFG of up to 10 for really crisp images, but I personally think 25 steps with a CFG of 7 is the sweet spot for a good balance. There really aren't many samplers that don't work well, so you should be good to use whatever is your favorite. I default to DPM++ 2S a Karras, but using other samplers doesn't degrade the image's quality.
More about Version 6
Version 6.0 is a revolution on the standard Deep Dark Hentai Mix. It can do pretty much anything—humans, animals, furries, mo, females, males, futa, and pretty much any combination of all of the above. Prompting is more or less straightforward, though certain and specific concepts may only work with either natural language or Danbooru tags depending on how they were trained into the model and which component model understood them. Version 6.0 wouldn't have been possible without the collaboration with DucHaiten, so check out his stuff and consider supporting him. Read more under version details -->
Deep Dark Hentai Mix is an SDXL model geared toward anime and hentai image creation. It's capable of both “safe” generations and NSFW soft and hardcore generations. See the Usage Tips and Prompting section if your images are not coming out right.
I hadn’t been able to find an SDXL anime model with the exact style and aesthetic of some of my favorite 1.5 models, so I set out to combine merging and training to achieve something similar for SDXL. The results of the first attempts produced only images of the deepest, darkest horrors from the abyss. However, after a few tries, I achieved something that was semi-stable, produced nice quality images with some well-thought-out prompts, and had the look I wanted. The only problem was that it was also extremely horny. It removed clothes, interpreted every prompt in the most sexual way possible (and in some cases in ways that were not possible), and generally created some of the most lewd, disturbing, and horny images possible. While trying to fix the issue, I accidentally merged it with some of the earlier failed deep dark model attempts and found that, oddly, the new merge both somewhat cooled things down (so it mostly only produces hardcore content on request) and was a mostly stable model. Thus, the Deep Dark Hentai Mix was born.
For anyone curious, I released that very horny, crazy model on its own page as Hentai ProtoGenesis XL.
Model Versions
Unlike other models, major versions of Deep Dark Hentai Mix (e.g., version 1.0 vs version 2.0) represent different lineages and methods of merging. If you like the style and capabilities of a particular version number, check out the newest model in that version. Details about each version are to the right under “About this version,” but in general, the differences are:
Version 1 Models: All Version 1 models are made with the same merging technique and recipe, which uses an odd combination of model weights from realistic and anime models that produce the Deep Dark style. It basically involves training realistic models to default to an anime style. New releases in Version 1 use the most up-to-date models (both my own unreleased models and other publicly released models made by other creators).
Version 2 Models: Models from Version 2 were created using an alternative recipe based on Version 1 and tend to produce more semi-realistic and 3D art styles by default.
Version 4 Models: A new merging technique that replicated much of the style from earlier models but has an increased capability to understand and follow NSFW prompts.
A note on watermarks: Yep, it's got them. Some versions have significantly fewer watermarks; later versions can usually make a batch without them. I've managed to get it down to about 1 in 10 images with prompting. Check below for usage tips for more information on how to limit them.
Usage Tips and Prompting:
Most versions of the model are very picky about sampler settings. I typically use any of the following settings, all of which work great: Euler a, DPM++ 2S a Karras, DPM++ 3M SDE Exponential, and DPM++ 2M SDE Karras.
This model generates quality pictures at very low steps – 25 steps will work fine. If you want super quality, 50 steps seems to be the sweet spot.
I don't use a refiner and only use hiresfix when upscaling to get a larger image. No detailers or inpainting. The point was to make a model that just made great images out of the box.
Give it a low CFG so it has the freedom to create some really interesting things. I typically use between 6 and 9, but I have gone as low as 3 and as high as 12 or 13 before images start looking burned.
Natural language works best for semi-realistic images. Danbooru tags will produce mostly 2D anime-style art. All of the versions of the Deep Dark Hentai Mix model will mostly produce anime-style images when prompted with Danbooru tags (1girl, 1boy, etc.) but will produce fantasy art and semi-realistic styles for some prompts in natural language. It knows certain tags from particular styles, so many times it will switch styles when prompted with a particular anime trope. It can do SFW images, but really doesn’t like to. It really prefers mature, adult content and will happily produce a wide range of softcore and hardcore images. Without careful prompting, pants will be missing, anatomy will be on display, things will be inserted, etc. You’ve been warned.
If you want to improve the quality and resolution of images, use tags like high resolution, highres, best quality, 4k. Intricately detailed and amazing details work to add extra complexity. Cinematic pushes images to 3D and makes some generations realistic. Don't use masterpiece, immaculate, absurdres, absurd res, 85mm -- they all do more harm than good. No negatives are needed as a standard prompt; things like low quality or bad anatomy aren't necessary.
Some of the models included in this merge are capable of producing some really beautiful artwork. Even if you aren’t into hentai, much of the time you will be able to generate some really great SFW images with very good anatomical details.
Watermarks come and go, but they are obviously tied to certain keywords (which may differ between versions), so limiting certain words will significantly cut down on their presence. I don’t know why, but I see about 90% less watermarking when I use this addition to my prompts:
Added at end of the prompt: NSFW, erotic, perfect body, sexy, 20-years-old; added at end of the negative prompt: SFW, children, loli, child, kids, teens
This model is not intended and does not tend to produce illegal acts and images. However, to be 100% safe, add "children, loli, child, kids, teens" to your negative prompt. The model understands "girl" = "woman," but using the latter is safer. 1girl is usually safe for anime images, but I’d still include a negative prompt.
Model Credits
The following models have been or are currently used in at least one version of Deep Dark Hentai Mix. Credit goes to the model creators (and, of course, the creators of the models used in their merges/training). In addition, a number of trained models or trained LoRAs may have also been used that I haven't released for one reason or another.
DucHaiten-Infinity-Pony https://civitai.com/models/412724/duchaiten-infinity-pony
Trifecta XL https://civitai.com/models/286936/trifecta-xl
Hentai ProtoGenesis https://civitai.com/models/244551?modelVersionId=275930
ShikiAnimeXL https://civitai.com/models/119163/shikianimexl
Hentai Mix XL - Road to Freedom https://civitai.com/models/199554?modelVersionId=296678
Hentai Mix XL - World is flat! - https://civitai.com/models/289879/hentai-mix-xl-world-is-flat
Aesthetic Mix XL https://civitai.com/models/262187/aesthetic-mix-xl
SDXL Nuclear - General Purpose v2 https://civitai.com/models/138331?modelVersionId=162051
NSFW POV All In One SDXL LoRA https://civitai.com/models/144203?modelVersionId=160240
Deep Dark Hentai Mix https://civitai.com/models/221751?modelVersionId=256648
Pyro's NSFW SDXL https://civitai.com/models/133808/pyros-nsfw-sdxl
duchaitenReal3dNSFW_v10 - https://civitai.com/models/247266/duchaiten-real3d-nsfw-xl?modelVersionId=286445
omnigenxlNSFWSFW_v10 https://civitai.com/models/203014/omnigenxl-nsfw-and-sfw
betterThanWords_v30 https://civitai.com/models/149359?modelVersionId=233092
FappXL LyCORIS https://civitai.com/models/248856/fappxl-multi-nsfw-concept
And a few other models that have vanished now and are no longer posted on model cards.
Legal Stuff
AI model output may sometimes be unpredictable and is not guaranteed to match the input prompt nor to be of any given nature or quality. Do not attempt to generate content that is illegal in your legal jurisdiction or within your country or district of residence. By using this model, you agree that the creator of this model shall in no event be responsible or liable to you or to any third party, whether under contract, warranty, tort (including negligence), strict liability, indemnity or other theory, for any direct, indirect, special, incidental, consequential, exemplary, liquidated or punitive damages or any other damages, including but not limited to loss of profit, revenue or business, cost of substitute procurement, arising in whole or in part from your use of (or inability to use) the model.
Model Restrictions:
To be clear, this model is a derivative of the SDXL 1.0 model and the following restrictions are part of the 2023 Stability AI CreativeML Open RAIL++-M License dated July 26, 2023, with applies to this model and all models that use any component or derivatives of this model. You agree not to use the Model or Derivatives of the Model:
In any way that violates any applicable national, federal, state, local or international law or regulation;
For the purpose of exploiting, harming or attempting to exploit or harm minors in any way;
To generate or disseminate verifiably false information and/or content with the purpose of harming others;
To generate or disseminate personal identifiable information that can be used to harm an individual;
To defame, disparage or otherwise harass others;
For fully automated decision making that adversely impacts an individual’s legal rights or otherwise creates or modifies a binding, enforceable obligation;
For any use intended to or which has the effect of discriminating against or harming individuals or groups based on online or offline social behavior or known or predicted personal or personality characteristics;
To exploit any of the vulnerabilities of a specific group of persons based on their age, social, physical or mental characteristics, in order to materially distort the behavior of a person pertaining to that group in a manner that causes or is likely to cause that person or another person physical or psychological harm; For any use intended to or which has the effect of discriminating against individuals or groups based on legally protected characteristics or categories;
To provide medical advice and medical results interpretation;
To generate or disseminate information for the purpose to be used for administration of justice, law enforcement, immigration or asylum processes, such as predicting an individual will commit fraud/crime commitment (e.g. by text profiling, drawing causal relationships between assertions made in documents, indiscriminate and arbitrarily-targeted use).
Clarifications on the permissible use of this model:
This model is a merge. Therefore it inherits the permissions and restrictions of all models used in its creation.
You may not make merges, extractions, or derivatives of this model and use a less restrictive license than the 2023 Stability AI CreativeML Open RAIL++-M License.
This model may be run and hosted for download on Civitai.com. At this time, no other organization or individual is authorized to host or distribute this model.
You may not sell this model or any of this model’s merges or derivatives.
You must credit this model’s creator if you use this model to create a new merged, mixed, trained, or extracted model.
You may sell images you generate with this model.
You may share merges using this model as long as any merge has the same or more restrictive permissions.