<h1 id="zylon-the-alien-character-lora-flux-7j4d7x4gx"><strong><u>Zylon The Alien - Character Lora - Flux</u></strong></h1><p></p><h3 id="key-details:-zohea7z2c">Key Details:</h3><ul><li><p>Lora strength: 1-1.5</p></li><li><p>Keywords: zylon, alien, dark medallic skin</p></li></ul><p></p><h3 id="further-details:-4w6b3qol2">Further Details:</h3><p>This is an experiment for a character lora workflow. The idea was to:</p><ol><li><p>Have an ai write an empty character sheet with details about age, appearance, and a full backstory.</p></li><li><p>Have the ai fill out that template with a simple prompt of "fill out the sheet for an alien called zylon"</p></li><li><p>tell the ai to "now wright this in to a paragraph description of zylon"</p></li><li><p>Generate 4 images using that same description with different seeds using flux</p></li><li><p>Used my ltx looped video workflow to create 4 videos from the images, you could use any img2vid workflow for this and will probably get the best results from kling.</p></li><li><p>wrote a quick python script to split the video in to frames grabbing and saving every 10th frame from the video.</p></li><li><p>trained a flux lora off of the images.</p></li></ol><p></p><p>From quick testing the results seem to be pretty decent. I've included the training data of the images I used if it's of any use to anyone. I will be attempting this workflow idea again a couple times to see if it's a one-off fluke or if it's a reliable way of getting training data for character loras.</p><p></p><h1 id="-13wp3r3rk"></h1>

Files with SHA256: 4edd67e785db902c34e84f9dea832abbf265ceaffb622395e6c72da909fa3414