<h1 id="this-checkpoint-is-under-construction!">This Checkpoint is under construction!</h1>Please read this post 😌<hr />I wrote an article about this model and the new strategy for captioning, <a target="_blank" rel="ugc" href="https://civarchive.com/articles/37878/llms-as-text-encoders-the-benefits-the-costs-and-a-new-approach-to-training">click here to check it out.</a><hr /><h2 id="all-sample-images-are-zero-shot-generated">All sample images are zero-shot generated</h2>I will always try to avoid prompts that are similar to those used in the dataset. This way, I can show you the progression and quality of learning, making it clear that the model is learning the concepts and not just memorizing the images.Even though no faces were blurred or removed from the images in dataset, you'll still see that it will always generate random faces (you won't see Mia Khalifa or Kendra Lust in the generated images 😌)<hr /><h2 id="prompt-upsampling">Prompt upsampling</h2>The sample images were not upsampled; the prompt was passed directly to the main text encoder of Qwen 2.1 Turbo. The clever approach of using a dedicated text encoder solely for upsampling complicates the training process somewhat, as you must use captions that resemble the output the PUP (Prompt Upsampler) will generate. I decided to perform a fine-tuning run that works for the base model while remaining compatible with any PUP. I have been using Mistral Small and Command R Reasoning as PUPs with excellent results; I will soon post images generated using a PUP, showing both the original prompt and the detailed, upsampled prompt.<hr /><h2 id="known-issues">Known issues</h2>LOSS mask images leaked into the training data; consequently, if the prompt is not understood, the model may generate a gray image.<hr /><h3 id="the-images-generated-by-this-model-are-not-at-an-acceptable-level-so-don't-expect-extraordinary-results-(yet).">The images generated by this model are not at an acceptable level, so don't expect extraordinary results (yet).</h3>❌ = Not available yet🔴 = Worst🟠 = Not good🟡 = Almost there🟢 = Good🟣 = Perfect<hr /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/dfec5a73-3478-46d0-bc0c-fbab3000af7d/original=true/dfec5a73-3478-46d0-bc0c-fbab3000af7d.jpeg" /><hr /><h1 id="available-in-v5.0-dataset-(initial-validation)">Available in v5.0 dataset (initial validation)</h1><h3 id="concepts-implemented-on-this-version:">Concepts implemented on this version:</h3><ol><li>Anus 🟢</li><li>Back view 🟢</li><li>Breasts size control (small, medium and big) 🟡</li><li>Black penis 🟢</li><li>Blowjob 🟡</li><li>Cum in mouth 🟠</li><li>Cum on face 🟡</li><li>Pussies 🟢</li><li>Doggy-style (dedicated) 🟡</li><li>Fingering ❌</li><li>General sex positions (Missionary, Doggy, Spreading, POV, etc) 🟡</li><li>Groping ❌</li><li>Hard nipples (need more images) 🟠</li><li>Naked man 🟢</li><li>Panties aside (need more images) 🟠</li><li>Playing with pussy (Rubbing, touching, etc) 🔴</li><li>POV (dedicated) 🔴</li><li>Pussy closeup 🟢</li><li>Removing clothes 🟠</li><li>See through (need more images) 🟢</li><li>Penis control (flaccid, hard, small, medium, big, thin, thick, etc) 🟢</li><li>Text 🟢</li><li>Weird things (Funny and weird images, like "stepsister stucked in washing machine" 😏) 🟠</li><li>Wet clothing (need more images) 🟢</li><li>Female masturbation (hands inside shorts, panties, skirts, etc) 🔴</li><li>African women (need more images) 🟢</li><li>Normalization (simple prompts for complex things) ❌ (it is no longer necessary because of the prompt upsampler)</li></ol><hr /><h3 id="concepts-to-be-incorporated-soon">Concepts to be incorporated soon</h3><ol><li>Sex with panties aside (dedicated)</li><li>Big breasts (to improve breasts-size control)</li><li>Sexy clothing</li><li>E-girls</li><li>Cosplay</li><li>Amateur Selfie</li><li>Wide hips (with narrow waist)</li><li>Chubby women</li><li>Fat women</li><li>Interracial sex (dedicated)</li><li>Femboys and Tomboys (dedicated)</li><li>Public (dedicated)</li><li>MILF</li><li>Lesbian (dedicated)</li><li>Teasing</li><li>Gangbang</li><li>Footjob</li><li>Cameltoe</li><li>Bathroom (dedicated)</li><li>Asian (dedicated)</li><li>Nature (dedicated)</li><li>Gay (dedicated)</li><li>Spanking/Discipline</li></ol>Items marked as "dedicated" are already present in the dataset, but more images are needed for better learning.<hr /><h2 id="how-about-helping-to-make-the-dataset">How about helping to make the dataset?</h2>There are many concepts and fetishes, and I don't know all of them and may not even be able to remember most of them, so you can suggest a concept to be included in the model.You can just suggest the concept or send images/links of the concept to my <a target="_blank" rel="ugc" href="https://huggingface.co/Aisha-AI-Official/LoRA-requests/discussions/16">HuggingFace on LoRA Requests community</a>.<hr /><h2 id="the-checkpoint-purpose:">The Checkpoint Purpose:</h2>I'm trying to bake the best SFW + NSFW Base Model.I focused heavily on regularization so that the NSFW wouldn't end up leaking and contaminating the native SFW side of Qwen. To do this, I needed to create a large regularization dataset, which significantly delays training, making it more expensive 😥And of course, I'm not adding knowledge, I'm replacing knowledge that is less used to make room for NSFW.My intention is to make this model incapable of producing anime-style images, so we will have "more space" for realistic NSFW content and prevent Qwen from forgetting how to do other things, such as writing text and editing images (called "catastrophic forgetting").<hr /><h1 id="want-to-help-me-continue">Want to help me continue?</h1>Buy PaidCoins at <a target="_blank" rel="ugc" href="http://Aisha-AI.com">Aisha-AI.com </a>and sent them to account number 2orCreate an account on <a target="_blank" rel="ugc" href="https://cloud.vast.ai/?ref_id=212422">Vast.ai</a> using my referral code (<a target="_blank" rel="ugc" href="https://cloud.vast.ai/?ref_id=212422">https://cloud.vast.ai/?ref_id=212422</a>), top up your account, use the option to transfer the balance to another user and provide my ID: 212422<hr /><h3 id="total-spent-on-this-checkpoint:">Total spent on this Checkpoint:</h3>$110 USD (😐)

Aisha 💜 NSFW [Qwen 2.1 Turbo]

First v5 dataset validation.This checkpoint serves as a validation test for the new captioning style to be used, which relies on a new pipeline featuring a prompt upsampler. The test was a success and has begun yielding good results. The v5 dataset is now being assembled using this new Markdown-based captioning standard. For more information, please <a target="_blank" rel="ugc" href="https://civitai.red/articles/37878/llms-as-text-encoders-the-benefits-the-costs-and-a-new-approach-to-training">refer to this article.</a>

Files with SHA256: df6fc5b53e337d7dc97cb6424ee42f82f2c6e06d2d927f747595a73b3f37fe49