<p><span style="color:rgb(121, 80, 242)">If you like the model, please leave a comment or post an image! 🥰<br />Or if the model doesn't work as well as you expect, let me know with a comment! 💬 I can always add more training data.</span></p><p>A LoRA that replaces basically every other metal bondage frame LoRA because it does ALL THE POSES. If you can prompt a character to do a pose, this lora will put them in a bondage frame for that pose!<br />If this doesn't work for the pose you want, <s>you get your money back!</s> you should leave a comment so I can add it to the training data.</p><p><span style="color:rgb(253, 126, 20)">V2 has been tested with the turbo model checkpoint!</span></p><h1 id="v2:">V2:</h1><p>V2 has much better control over the frame types and frame coherence. It also works way better with the Anima Turbo model! (From my testing). The style bleed is a bit worse though and camera angle diversity might be slightly worse. 😅 </p><h3 id="prompting">Prompting</h3><p>Mostly the same with the frame types renamed for better concept isolation.</p><p><code>[pose] in a [frame type]</code></p><ul><li><p><code>[pose]</code> - same as V1, almost any pose should work</p></li><li><p>[frame type] - <code>bondage frame</code> or <code>metal vine bondage</code> or <code>metal harness</code></p><ul><li><p><code>bondage frame</code> - was called <code>loose frame metalbond</code> in v1. Rigid rectangular metal frames.</p></li><li><p><code>metal vine bondage</code> - was called <code>tight metalbond</code> in v1. Rounded/curvy metal shaped to the character's body. Artistic.</p></li><li><p><code>metal harness</code> - Really shaped to the character's body, like a harness made of metal.</p></li></ul></li></ul><p></p><h1 id="v1:">V1:</h1><h3 id="prompting">Prompting</h3><p><code>[type of] metalbond [attachments]</code></p><ul><li><p><code>[type of]</code> is primarily either <code>loose frame</code> or <code>tight</code> and determines how tightly the metal wraps around the girl</p><ul><li><p><code>no frame</code> also helps increase tightness, but can sometimes remove necessary parts</p></li></ul></li><li><p><code>[attachments]</code> - use <code>suspended by chains</code> for hanging by a chain, <code>with a metal base</code> for a solid metal base on the ground, <code>with handle bars and wheels</code> to make it a hand truck (look it up)(doesn't work well)</p></li></ul><p>as always, prompting with similar words will help. use terms like <code>metal collar</code> or <code>wrist cuffs connected to frame</code> or <code>metal pole belt</code> to help.</p><h3 id="poses">Poses</h3><p>Just check the attached XY plot image.<br />I know I said any pose (and it some poses don't work well let me know!) but there are specific poses tested to confirm model effectiveness.</p><ul><li><p><code>spread arms and hands up and crouching spread legs</code> - pose in the shape of "<strong>H</strong>" with arms and legs at 90 degree angles</p></li><li><p><code>bowlegged</code> - knees apart feet together</p></li><li><p><code>legs together</code></p></li><li><p><code>arms by side</code> - often ends up more like <code>arms behind back</code></p></li><li><p><code>arms behind back</code></p></li><li><p><code>folded with legs up and arms down</code> - if you don't add <code>arms down</code> they can bleed into the legs</p></li></ul><h1 id="misc">Misc</h1><h3 id="img2img">Img2Img</h3><p>This model works really well with drawing solid grey lines and turning them into metal bars around the character, so if prompting doesn't get the exact form you are looking for, drawing them on can really help!</p><h3 id="thoughts">Thoughts</h3><p>Synthetic data was the real king in making this LoRA effective, and was why the v2 should work so much better for making coherent frames. But its also responsible for the style bleed and similar camera angles. That's my bad. </p><p>Why did the turbo model have to come out right after I released ths model? took me a bit of testing to determine why it had mixed results, and I think turbo tends to gravitate more strongly to average or strong features in the lora. So if you have any concept or style bleed those tend to become much more pronounced. So it still "works" but is less flexible.</p><p></p>

Files with SHA256: d8436174a5556ab6a8062e8e3ab574e306cdd983020313734da931806c8883e5