<h1 id="illustrious-cutting-t-shirt-challenge-l0553e2n2">[Illustrious] Cutting T-shirt Challenge</h1><h2 id="description-l1ipwgtm8">Description</h2><ul><li>Challenge that viral in X/Twitter</li><li>Work in progress LoRA (maybe some edit dataset to produce better LoRA)</li></ul><h2 id="trigger-word-shi25vbuc">trigger word</h2><pre><code>cutting_challenge, like and retweet, number, white shirt</code></pre><h3 id="limitations-ixhhx164b">Limitations</h3><ul><li>Number of favorite and retweet is not accourate</li><li>Text stills not good</li></ul><h2 id="training-details-ez92vc0fi">Training Details</h2>LoRA rank<ul><li>LoRA standard</li></ul>dataset<ul><li>72 images</li></ul>parameters<ul><li>resolution = 1024</li><li>batch size = 2</li><li>network dim,alpha = 12,12</li><li>mix/save precision = bf16/bf16</li><li>optmizer = Lion (weight_decay=0.05)</li><li>UNet LR = 3e-5</li><li>TE LR = 1e-5</li><li>scheduler = cosine_with_restarts (3 restarts)</li><li>l2 loss</li><li>alpha_mask</li></ul>steps<ul><li>epochs = 10</li><li>total steps = 3600</li><li>repeat = 10</li></ul>tools<ul><li>kohya-ss GUI v24.3.0 FORKED by vjumpkung</li><li>torch 2.5.0 cu124</li><li>Runpod RTX 4090</li></ul>avg weight<ul><li>UNet average weights : 0.009967144748026674</li><li>TE1 average weights : 0.007946787285618484</li><li>TE2 average weights : 0.006517791320220567</li></ul>*This LoRA is for studying LoRA training with new technique so do not use for challenge in X/Twitter.*If you see my model reposted on other image generation service please report to me.

[Illustrious] Cutting T-shirt Challenge

test version with remove background dataset

Files with SHA256: 411218b8b2f7e54ecdc760efc4f456f92045bb07b716e5a7b497969eade8a621