This model will help you draw male/male intimate encounters.
The model follows an ambiguous prison setting, meaning that this setting tends to not be overtly apparent unless you choose to lean into it through prompting. Adding a dynamic and immersive prison setting will require a stand-alone LoRA. Stay tunned for that, it's in the pipeline.
Trigger: zpipeline
This project marks a departure from my previous training style and philosophy. "Secondary Triggers" are no longer necessary. Use the trigger word zpipeline and then prompt whatever you might want in your image. The model will create your prompt through the lens of the concept/s that it has learned, those concepts being related to male/male intimate encounters. I will add prompt ideas that are relevant to the concept.
Positive Prompt Ideas:
yaoi - bara - inmate - prisoner - white_t-shirt - green_pants - prison_cell - prison_shower - ???? - rough_sex - deep_penetration - sex_from_behind - standing_sex - bent_over - elbows_on_table - elbows_on_bed - arched_back - lying, on_bed, on_stomach - prone_bone - on_side - spooning - on_back - missionary - legs_up - knees_to_chest - kneeling - doggystyle - male_spitroast - mmm_threesome -???????? - irrumatio - facial - tongue_out - ????????? - bullying - wince - grimace - moaning - foreboding - emotional - intense
*One prompt strategy would be to structure the prompt similar to this example -
yaoi, "add a few, concise, natural language prompts describing what you would like the image to depict", LoRA tag, no comma, space, trigger word, comma, "a series of tokens or prompts that add specific elements and or reinforce your natural language description and or the concept of this model" -
yaoi, there are two inmates in a prison_cell getting to know each other, they are ?????????? each other as a matter of politeness, <lora:Pipeline2:1> zpipeline, white_t-shirt, green_pants, undressing_another, erection_under_clothes, nervous, shy, excited, implied_yaoi, intense, 
Negative Prompt Ideas
1girl - ????? - breasts - twins - hetero - bad ??????? - worstquality - out of focus - blurry - low quality - bad proportions - mutated - poorly drawn face - extra limb - mutation - ugly - disfigured - deformed - bad body - missing fingers - long body - size_difference
*It may be a good strategy to avoid using negative prompts preemptively. They could potentially create a problem in your image where one would not have otherwise materialized. Use negative prompts as a reaction to counter things that you do not like in the images you have previously generated in your current workflow.
The example images for v2 were created with the following checkpoints -
NTR MIX | illustrious-XL | Noob-XL
SDXL VAE
********************************************************************************************************Note the width/height as they are not a standard size. 1216x832 is not an aesthetic ratio, so I bump 832 up to 896 which gives images a more pleasing ratio of approximately 3:4. I don't know if using my models with other dimensions, apart from the three I list below, warps images because I only ever use 1:1 or 3:4/4:3.
The following section contains information that is merely a suggestion for those whom might be new to image generation or just looking for some guidance. Many other settings for these parameters will also work.
Clip Skip: 1
CFG: 3-15
LoRA Strength: 1.0 (+/-)
Sampling Method: Euler - DPM++ SDE - DPM++ 2M
Schedule type: Karras
Sampling Steps: 20-45 (+/-)
Width/Height: 1024x1024, 1216x896, 896x1216
********************************************************************************************************This is a concept LoRA that has been trained on images of anatomically correct males of various ethnicities, ages, and body types. All images used in the data set were my own original artwork and were generated using Stable Diffusion SDXL. (Not the base model itself, but a checkpoint that was fine tuned from the SDXL base model.) This is noted because SDXL LoRAs work better when they are trained on SDXL generated images. Think of it like the training to create the LoRA being in a language Stable Diffusion is already fluent in.
Description
This LoRA creates realistic depictions of incarcerated young men engaging in various ??? acts with eachother. It was created with 127 original stable diffusion images of my making at about 10k steps. All images in the data set were auto captioned and then personally pruned and amended. The captioning is tag-style using danbooru tags. That means many of the tags contain underscores, ie. - "sex_from_behind" will probably get you further than prompting "??? from behind".
The trigger is "prx_concept". This model will draw anatomically correct and aesthetically pleasing penises and erections of various sizes, granted that you are using a checkpoint that also knows how to draw them or are using an additional LoRA/ect to help guide the magic towards a human Earth-????, as opposed to the dystopian nightmare ????? that live inside the majority of models.
Not specifying a racial description of your subject/s tends to draw Caucasians. "dark_skinned" or "dark_skinned_male" tends to draw African-Americans. "latino" will conjure Latinos.
Stable diffusion has many biases, as it can only know what it has aleady been taught. One of these biases is against pubic hair. The tag "excessive_pubic_hair" will usually get you neat and trim pubic hair. Not using it will usually result in a shaved/waxed situation.
The theme of this LoRA is more focused on the social interactions and behaviors of the inmates rather than the physical setting of a prison. The following list are prompts that can help increase the chances of generating prison themed backgrounds:
"prison, inmate, prison_guard, cell, prison_cell, cellmate, prison_clothes,prison_bed, prison_shower, prison_storage_room"
There's a rumour going around on the yard that some of the inmates may have stolen a knife from the kitchen, or perhaps fashioned a makeshift equivalent.
The model was trained on images of various sizes that corresond to standard SDXL dimensions and should do well in either landscape or portrait.
The reference images I provided are all unaltered and come straight from text to image using this LoRA in prompts that did not exceed 75 tokens. No other LoRA or textual inversion, ect, were used in their creation. DPM++ 2M KARRAS, 38 STEPS
*I originally uploaded four images using the criteria explained directly above this clarification. I have/am adding supplemental images showcasing this model that were/will be subject to a more relaxed criteria. I will separate these images into subcategories if possible.
Negative Prompts are not neccessary but "?????" and "1girl" are helpful as negative prompts if you see mangina penetrations materializing. That should get things keistering properly.
There are many danbooru tags the model will respond to and even a few non-danbooru tags of plain English words that I taught the model that are related to the subject matter. Maybe you'll get lucky and discover a few along the way.
Please post some images on this page of any particularly spicy or interesting things you might create using this resource.
Enjoy.
FAQ
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.