<h1 id="caption-and-token-quality-comparison-jxmbvifcn">Caption &amp; Token Quality Comparison</h1>The Lora compares three onsite trainings all using the same dataset and training settings with captions described in the study below.<ul><li>Default Pony Settings used other then batch size 4</li><li>10 repeats per image 10 epochs</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/api/download/models/1639625?type=Training%20Data">Dataset</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/api/download/models/1640017?type=Training%20Data">Caption Data</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/articles/13444/summerize-onsite-captions-simple-gui">T5 Online Caption Summary Tool</a></li></ul><h3 id="study:-ak7wg3owt">Study:</h3>Training<ul><li>50 Royalty Free images have been captioned onsite, and trained. LORA Big ASP.</li><li>The Same images and captions trained with the same settings but using the T5 summary tool. LORA T5</li><li>Again the same images but with captions made by hand. LORA Hand-Captioned</li><li>All Training done on base PONY using Captions setting not TAG</li></ul>Image generation<ul><li>Images are seed to seed comparison with CyberRealistic Base model</li></ul><hr />Royalty Free images from 2008

Caption & Token Quality Comparison

Files with SHA256: 8fdd0da58f4fd7c01907a7ca9b4045ffa44d3f4a898102dcb754855f50614db3