THTU Beta V2 Changelog:

Prompt Structure Changes: I've updated the prompt structure significantly. You no longer strictly need to use clinical terms like 'buttocks' or 'vulva'. The model now perfectly understands and reacts to words like "ass", "pussy" and "asshole" just the same.

Training Adjustment: I tweaked the training settings, resulting in a visible reduction in anatomical glitches and action failures. You should now be able to achieve your desired composition with a much higher success rate.

The full version of the LoRa will require a significant number of front-facing photos and a wide variety of clothing interactions (specifically panties being pulled aside, twisted, etc.). Because gathering high-quality data for these specific interactions takes time and because I want to experiment with a few different concepts, the Full Version will be delayed a bit.

That's why I wanted to release this second Beta to provide a better, smoother experience in the meantime. Thank you for your interest!

Quick Prompting Guide:

Essential Tags: Due to the base model's bias towards generating front-facing subjects, prompts like "taken from behind" and "facing away from the camera" are practically mandatory.

Action Trigger: It seems to work even without describing specific clothing if you use phrases like "expose ass, pussy, asshole" within your desired sentence structure.

Recommended Actions: Standing, Leaning, Bending, Kneeling, All fours.

Recommended Clothing Actions: Lifted, Pulled down/up, Bunched up.

Even if you don't use these exact words, the model usually understands your intent and tries to execute it. These are simply the words I relied on during captioning for training, so they are the most reliable.

Flexibility: Beyond that, feel free to describe the subject or the environment in as much detail as you like; this LoRa does not restrict composition creation. I see no reason why it shouldn't work with other LoRas; you might just need to lower the weight. I've seen it work down to 0.6 in some compositions.

Note: As the prompt becomes more detailed and complex, the success rate of the specific action may drop. If you notice that none of your consecutive attempts are working, please try changing your sentence structure.

Too Horny To Undress [THTU]:

The concept is simple: Ladies who are just too horny to bother taking their clothes off, getting 'ready' in the most optimal way possible. Yeah... Simple...

After the experience gained from my Flashing Tits Lora, I thought it made sense to create a 'Flashing Ass' version. But as things progressed, I realized that might be a bit boring on its own, so I pivoted the concept. This Lora isn't just about the act of flashing; it covers suggestive positions and brings a massive quality boost to areas like the vagina and anus.

You might say: "Hey Comrade, Klein already struggles with anatomy, refuses to do sexual acts, and is oblivious to half the human body. Is it logical to do this?"

My answer: "No, it's not logical. I haven't made a logical decision in the last 7-8 years of my life... but strangely enough, IT WORKS!"

Quick Prompt Guide:

Flux2 Klein relies heavily on natural language prompting, so traditional trigger words don't function quite the same way. You need to describe the composition clearly for the model to understand. Here is a logic you can follow:

Structure:
[Shot, Subject & Position] > [Clothing & Action] > [Environment details] > [Lighting & Mood]

Stage 1: Shot & Subject
Define the camera angle and the subject's pose. Since this lora focuses on the rear view:

• Template: A [shot type: close-up, full body, wide shot, POV] photograph taken from behind of a [subject description], [standing / bending over / kneeling / on all fours] in [setting]. She has [anything]

• Selection: "A high-angle, full-body photograph taken from behind a woman standing on a sandy beach. She has sun-bleached wavy hair and pale skin."

Stage 2: The Action (Crucial!)
Describe the clothing state and the exposure explicitly.

• Template: She is wearing [clothes], [action: pulled down to ankles(or thighs or knees/ lifted up / hiked up / flipped up], [exposing her bare buttocks / leaving her buttocks completely bare].

• Selection: "She is wearing an orange, cropped, sleeveless top and black, thigh-high leggings that are pulled down to her thighs, exposing her buttocks."

Stage 3 & 4: Environment & Mood (Optional)
Add the final touches to the scene.

• Selection: "The background features calm ocean waves under the warm glow of a setting sun, creating a soft and cinematic atmosphere."

FINAL FULL PROMPT EXAMPLE:

"A high-angle, full-body photograph taken from behind a woman standing on a sandy beach. She has sun-bleached wavy hair and pale skin. She is wearing an orange, cropped, sleeveless top and black, thigh-high leggings that are pulled down to her thighs, exposing her buttocks. The background features calm ocean waves under the warm glow of a setting sun, creating a soft and cinematic atmosphere."

General Guide:

• Check the Samples: Definitely check the prompts in the sample images. I trained this on hundreds of photos featuring actions like standing, leaning, kneeling, and all fours. The common thread is that they are all viewed from behind. So, prompts where the subject is facing away will work best. You'll see it works when you detail the "exposed" action the subject is doing. Feel free to use LLMs; Gemini is doing a good job right now for me.

• Why Beta?: The reason this is currently in 'Beta' is that I plan to conduct tests with some changes in my training methods, and of course, I have a dataset several times larger waiting to be used. I also plan to add front-facing shots in the next stage to capture a general "clothed horniness" vibe. I hope you share my taste for "quick" pleasures and that this excites you...

• Anatomy: Based on my initial observations, the vagina and anus give an 'acceptable' result about 50% of the time, unless it's an extreme close-up. Again, there are no edits in the sample photos. For very close shots, you might need to generate a few times until you get a result you like. But I can say with confidence: it definitely knows what an anus and vagina are, right now...
  Note: Please remember that this is NOT a 'vagina/anus enhancer' Lora and it won't work that way; it only serves to carry the necessary standard for the specific action.

• Camera Angles: Generally, "facing away from camera" works flawlessly. Prompts like "facing the camera over her right shoulder" do work, but they increase the risk of anatomical errors. Unfortunately, the Klein is very prone to anatomy glitches.

• Avoiding Problems: None of the images I uploaded are cherry-picked; most are just the result of a few attempts. Don't forget: the model has anatomical issues, so you might need to reroll to get the right hands/feet. It's also quite weak on subjects like car interiors, so getting that specific "inside the car" shot might take quite a few regenerations.

• Feet: Yeah, they are terrible. Sorry. Either avoid describing them or fix them with in-painting.

• Hands: Even though I used actions like "pulling panties aside," "spreading buttocks," or "gripping buttocks" in training, they still give terrible results. Unless you want to generate hundreds in bulk to find a good one or edit them later, I suggest avoiding these terms.

I hope you enjoy the Lora. I'm grateful for any feedback you provide. If you struggle with prompting, let me know, and I can prepare an LLM instruction guide for this Lora. Also, if there's demand, I can share the path I followed for training. Honestly, I wouldn't say no to any inspiration for a new Lora concept either.

Happy creating!