<h1 id="dhl-tape-dress">DHL Tape Dress</h1><h3 id="description">description</h3><ul><li>some short trend in last year or last two years</li></ul><h3 id="trigger-word">Trigger Word</h3><ul><li>DHLTapeCostume, <code>She/He</code> wears a tight-fitting strapless yellow dress printed repeatedly with the red “DHL” logo and the word “fragile,”</li></ul><h2 id="limitations">Limitations</h2><ul><li>None</li></ul><h3 id="training-details-v2">Training Details V2</h3>Dataset<ul><li>20 images (synthetic dataset)</li></ul>parameters<ul><li>resolution = 1024</li><li>batch size = 1</li><li>dim,alpha = 16,16</li><li>optmizer = AdamW</li><li>UNet LR = 1e-4</li><li>scheduler = cosine</li><li>timestep sampling : <code>flux_shift</code></li></ul>steps<ul><li>epochs = 5</li><li>total steps = 1500</li></ul>tools<ul><li>musubi-tuner</li><li>torch 2.8.0 cu128</li><li>Runpod RTX 5090 32 GB</li></ul><h3 id="training-details-v1">Training Details V1</h3>Dataset<ul><li>20 images</li></ul>parameters<ul><li>resolution = 768</li><li>batch size = 1</li><li>dim,alpha = 16,16</li><li>optmizer = AdamW</li><li>UNet LR = 1e-4</li><li>scheduler = cosine</li></ul>steps<ul><li>epochs = 5</li><li>total steps = 1500</li></ul>tools<ul><li>musubi-tuner</li><li>torch 2.8.0 cu128</li><li>Runpod RTX 5090 32 GB</li></ul>*This LoRA is for studying LoRA training with Z Image turbo*REPOST model is not allow if you found this model please report to me.

[Z-Image Turbo] DHL Tape Dress

<ul><li>change base model to Z-Image De-Turbo (by ostris)</li><li>major improvement by reduce training step and change resolution</li></ul>

Files with SHA256: 9548b6a300e49d344a6f94ed8c567dcf78d1d6d9f9c7f7c5e6be418b6d4d6911