<h1 id="version-0.4">Version 0.4</h1><p>Here is the updated version 0.4 for the plastic zip-cuffs.</p><h3 id="what-has-changed:">What has changed:</h3><p>- Triggerword is now "<strong>pla5ticz1pcuff5</strong>" (instead of "plasticzipcuffs") to make it absolutely unique. I had the feeling that the previous trigger word had already triggered existing content in the basic model.</p><p>- I reduced the size of the learning image dataset to minimize geometric distortions. This also explains the version jump from 0.1 to 0.4. The intermediate versions behaved too unpredictably.</p><h3 id="my-zipcuffs-look-strange-what-can-i-do">My zipcuffs look strange - what can I do?</h3><p>In my tests with version 0.4, the following procedure often helped:</p><p>- Start with strength 1.0, then vary the strength in the range of approximately 0.8 to 1.4 if needed. Sometimes smaller values ​​produce coherent images, sometimes larger ones.</p><p>- Change the seed.</p><p>- Deactivate other LoRas one by one to find out if there are any cross-influences.</p><p><strong>I admit that it is sometimes very time-consuming to obtain a realistic-looking image, especially when it comes to hand and finger positioning.</strong></p><p></p><h1 id="version-0.1">Version 0.1</h1><p>This model adapter was created by me to depict Ziptie restraints as used by the police.</p><p>It's not working as reliably as I'd hoped yet, but compared to the existing source material, it's very good. Often, however, several generations have to be initiated, as parts of the restraints can be missing or twisted. I hope to be able to take some additional photos of my own soon to generate a follow-up version, but I didn't want to withhold this from you. Perhaps someone will find it useful.</p>

Files with SHA256: b8c92022df751ee546ef2c43fdba683e14f7fafad260eadb25a8f44d3725d23e