<p>While Klein can already sort of do this, this lora reinforces the concept of a multi-view 'dressing' montage. <br /><br />Start with an input image of a nude person, and then describe their clothes - ideally start with a full body image, it's much less consistent if it's cropped. I purposely used non-cherry picked example outputs to give realistic expectations for the performance of the lora.</p><p></p><h2 id="i-strongly-recommend-generating-with-this-lora-using-this-output-resolution:">I STRONGLY RECOMMEND GENERATING WITH THIS LORA USING THIS OUTPUT RESOLUTION:<br /></h2><h2 id="1920-wide-x-1280-height"><span style="color:rgb(34, 139, 230)"><em>1920 wide x 1280 height</em></span></h2><p></p><h2 id="v1-v2-v3-differences:">V1-V2-V3 Differences:</h2><p>V1 - Best option if you have an input image that strictly adheres to the expected pose/distance</p><p>V2 - More flexible with input cropping / distance</p><p>V3 - More flexible with input image, I tried adding additional poses to the dataset<br /></p><p>I did not generate the input images, I just grabbed them and used them with the lora to generate the outputs!</p>

Dressing Sequence

<p>Modified dataset from V1 - meant to be more flexible with cropping of the input image<br />V1 is better if you have the strictly framed input image though</p>

Files with SHA256: 9321042d06b0a7451e6b8caebd57fb18b294ffbb0d111d1924c36dcf03879200