<h2 id="follow-me-on-patreon!"><a target="_blank" rel="ugc" href="https://www.patreon.com/Bitcrushed_Heart"><strong>Follow me on Patreon!</strong></a></h2><h1 id="soreal!-natural-bodies"><strong>SoReal! - Natural Bodies</strong></h1><p><a target="_blank" rel="ugc" href="https://civarchive.com/models/2360068">[SoReal! Portraits] </a><a rel="ugc" href="https://civarchive.com/models/2357375?modelVersionId=2657257">[SoReal POV]</a></p><h2 id="overview"><strong>Overview</strong></h2><p>Stand aside, supermodels! This model is the next iteration of my Z-Image LORAs - aimed to bring diversity in both concepts and humanity itself to Z-Image.</p><p></p><h2 id="compatibility-and-usage"><strong>Compatibility &amp; Usage</strong></h2><p>Due to it's small size and rank, the model should have a minimal influence on the base model, further improving compatibility with other LORAs across Base/Turbo and indeed other checkpoints.</p><p>'Trigger words' aren't real - don't ask for one, just prompt normally! Ages (xx-year-old format), weight classifications (underweight, natural weight, average weight, obese, etc), ethnicity and skin tones were used in captioning based on real data (rather than estimations). </p><p>When using Z-Image Turbo, strengths between 0.90 and 1.2 seem to have good results.</p><p></p><h2 id="limitations"><strong>Limitations</strong></h2><p>TBC</p><p></p><h2 id="future"><strong>Future</strong></h2><p><strong>I am planning on finetuning Z-Image considerably with a model called 'SoReal!' (Or, alternatively, ZoReal!). However, I want it to be the best possible amateur finetune possible, to achieve this, I have:</strong></p><ul><li><p>1. Trained a custom quality model.</p></li><li><p>2. Trained a custom one-shot demographic model (height, weight, skin tone, ethnicity, age in years, body shape) with an average accuracy of 89% for top-confidence prediction using ConvNext-XL.</p></li><li><p>3. Finetuned wd-tagger-large-v3 on a large sample dataset of 50k hand-tagged images with human-assisted active learning.</p></li><li><p>4. Fed those tagged images (with quality, demographics and general labels) with the image metadata (incl. EXIF &amp; Camera Metadata) to Gemini 3 Flash for generating captions.</p></li><li><p>No over-trained LORAs baked in, no dramatic loss of generalisation, just a good, all-round, NSFW-ready, finetuned model.</p></li></ul><p><strong>I am now severely limited, however, by my compute and financial situation, so if you'd like to help make SoReal!, well, so real, </strong><a target="_blank" rel="ugc" href="https://www.patreon.com/Bitcrushed_Heart"><strong>then you can follow me on Patreon!</strong></a></p><p></p><h2 id="dataset-and-training"><strong>Dataset &amp; Training</strong></h2><p>Dataset of 1000 sourced from a variety of sources. Deduplication and Quality Scoring (through MANIQA) batch size 16.</p><p>Validation loss was used with 10% of the dataset size to prevent overfitting while still maintaining strong concept adherence and generalisation.</p><p>Model was trained with AdamW through the Python adv-optm package.</p><p></p><h2 id="licensing"><strong>Licensing</strong></h2><p>If you'd like to release a merge of this model, please contact me.</p><p></p><p>Made with &lt;3 By BitcrushedHeart</p>

Files with SHA256: 163ffd4455ecdfd07d566455f3abb785c7513f2b0e828652d4b8fa17e8ea2344