Based on late 2010s era Honey Gold<h3 id="usage-tips-3y82dnelj">Usage Tips</h3><ul><li>Weight: 0.8-1</li><li>Trigger: <code>hngld</code><ul><li>The trigger word is optional. Simply mentioning "a woman", "she", or "her" seems to yield good results. Output can be unpredictable without any of the aforementioned terms.</li></ul></li></ul>A review/like/follow really helps and keeps me motivated! I'd love to see what you create and hear your feedback! ೕ(•̤̀ᵕ•̤́๑)ᵒᵏᵎᵎᵎᵎ<h3 id="training-info-w79i798fs">Training Info</h3><ul><li>Images: 60</li><li>Steps: 3000</li><li>Network Rank/Alpha: 32</li><li>Learning Rate: 1e-4</li><li>Tool: <a target="_blank" rel="ugc" href="https://github.com/ostris/ai-toolkit">ai-toolkit</a></li><li>Captioning: Florence-2-large-PromptGen v1.5, ChatGPT-4o, manual editing</li></ul><h3 id="misc-notes-2rn4u16q7">Misc Notes</h3><ul><li>Sample prompts and basic generation details are provided with the showcase images.</li><li>Showcase images have been upscaled (2x)</li><li>Captions were crafted in a natural language format, and with extensive details to enhance the LoRa's flexibility, covering aspects like hair, outfit, makeup, expression, pose, setting, and lighting.</li><li>A few issues have been observed during testing and are targeted for future updates:<ul><li>Her short, dark, asymmetrical hairstyle sometimes bleeds into the output when different hairstyles and colours are prompted.</li><li>Occasional generation of nudity/nipples when not prompted... (⁄ ⁄•⁄ω⁄•⁄ ⁄)⁄</li></ul></li><li>Tested using FLUX.1-dev</li></ul>

Files with SHA256: 7fed6c040e86031b7147bdb39fba89f0da859fb958ff0a518c79dc7d562cfea3