Upload 2357375/2691810/README.md with huggingface_hub

2357375/2691810/README.md

@@ -1,10 +1,114 @@
 ---
 license: other
 tags:
-- civitai
+- concept
+- nsfw
+- pov
 ---
+
+# SoReal! - POV - ZiB_V3 - 2691810
+
+**Model Type**: LORA
+
+**Base Model**: ZImageBase
+
+**Trigger Words**: None
+
+**Tags**: concept, nsfw, pov
+
+## Gallery
+
+<table>
+  <tr>
+    <td><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/0b538787-8fe2-4960-a014-1a21444ff36c/original=true/121178027.jpeg" width="200"></td>
+    <td><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/c89cd069-0c90-4aa9-a9ef-5f2c20b74ea5/original=true/121177996.jpeg" width="200"></td>
+    <td><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/fbc84350-0ae6-4327-b9de-ba18cd20af34/original=true/121178004.jpeg" width="200"></td>
+  </tr>
+  <tr>
+    <td><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/d9487bf2-31c5-408c-b451-5fc692a9228e/original=true/121179319.jpeg" width="200"></td>
+    <td><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/ccbc275a-f817-4cfd-9aef-031ecbc0bfa6/original=true/121178457.jpeg" width="200"></td>
+    <td><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/dad7b429-54e8-4cf3-9701-633a94087b0a/original=true/121178031.jpeg" width="200"></td>
+  </tr>
+  <tr>
+    <td><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/83d62144-e1e0-4a8a-96de-5ef868326ffa/original=true/121179058.jpeg" width="200"></td>
+    <td><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/a9a9208f-5759-4a96-b294-90bdbe243749/original=true/121178357.jpeg" width="200"></td>
+    <td><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/0569bbf6-fcf5-4318-9f4b-183ef1684c90/original=true/121178005.jpeg" width="200"></td>
+  </tr>
+  <tr>
+    <td><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/3d761615-4794-4606-9070-28ddaf64d28e/original=true/121179567.jpeg" width="200"></td>
+  </tr>
+</table>
+
+## Description
+
+[**Follow me on Patreon!**](https://www.patreon.com/Bitcrushed_Heart)
+---------------------------------------------------------------------
+
+SoReal! - POV
+=============
+
+[[SoReal! Portraits]](https://civitai.com/models/2360068)
+
+Overview
+--------
+
+Reach your hands out for the stars! This model is the first of a series in Z-Image LORAs - aimed to bring diversity in both concepts and humanity itself to Z-Image.
+
+Compatibility & Usage
+---------------------
+
+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.
+
+'Trigger words' aren't real - don't ask for one, just prompt normally. If you want a hand (literally), use 'a man's hand' or 'a woman's hand', which should normally get you what you want.
+
+I'll upload a full concept list soon to show the range of concepts the model has been trained on - not confirming that the model is able to reproduce them, though.
+
+When using Z-Image Turbo, strengths between 0.95 and 1.5 work best in my experience for V1, and 0.9 - 1.2 for V2.
+
+Limitations
+-----------
+
+Anatomy is still rough - planning one further additional training run to try and address this for NSFW concepts but may mean a split between generalisation model (v2) and a NSFW-special model (V2-NSFW).
+
+Future
+------
+
+Future iterations of this model will see stronger prompt adherence, anatomy adherence and general composition and quality through +/- reinforcement learning.
+
+**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:**
+
+* 1. Trained a custom quality model.
+* 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.
+* 3. Finetuned wd-tagger-large-v3 on a large sample dataset of 50k hand-tagged images with human-assisted active learning.
+* 4. Fed those tagged images (with quality, demographics and general labels) with the image metadata (incl. EXIF & Camera Metadata) to Gemini 3 Flash for generating captions.
+* No over-trained LORAs baked in, no dramatic loss of generalisation, just a good, all-round, NSFW-ready, finetuned model.
+
+**I am now severely limited, however, by my compute and financial situation, so if you'd like to help make SoReal!, well, so real,** [**then you can follow me on Patreon!**](https://www.patreon.com/Bitcrushed_Heart)
+
+Dataset & Training
+------------------
+
+Dataset of 2500 sourced from a variety of sources. Deduplication and Quality Scoring (through MANIQA) lowered the dataset to around 1400. This model was trained on a dataset of 1500 images at a batch size minimum of 10. This means
+
+This model was trained on a dataset of 1500 images at a batch size minimum of 10. Masked loss was implemented after roughly 40,000 samples (not steps) to improve anatomy & concept adherence.
+
+Validation loss was used with 10% of the dataset size to prevent overfitting while still maintaining strong concept adherence and generalisation.
+
+Model was trained with AdamW through the Python adv-optm package.
+
+Licensing
+---------
+
+If you'd like to release a merge of this model, please contact me.
+
+Made with <3 By BitcrushedHeart
+
+---
+
 Author: [BitcrushedHeart](https://civitai.com/user/BitcrushedHeart)
 
-Model: [https://civitai.com/models/2357375?modelVersionId=2691810](https://civitai.com/models/2357375?modelVersionId=2691810)
+Model: [CivitAI Model Page](https://civitai.com/models/2357375?modelVersionId=2691810)
+
+Archive: [CivArchive Page](https://civarchive.com/models/2357375?modelVersionId=2691810)
 
-Mirror: [https://civarchive.com/models/2357375?modelVersionId=2691810](https://civarchive.com/models/2357375?modelVersionId=2691810)
+<!-- Version: 20260422_v4_update -->