If you like the model, please leave a comment or post an image! 🥰 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.<h2 id="conditioned-model-is-img2img-only!">CONDITIONED MODEL IS img2img only!</h2><h3 id="use-v1.2-for-general-use">Use v1.2 for general use</h3>Please read before using the Conditioned Model:I made the conditioned model as a test using the same Cosmos reference latent method as seen in LoRAs like Anima Edit: <a target="_blank" rel="ugc" href="https://civarchive.com/models/2650553/anima-edit?modelVersionId=3089149">https://civarchive.com/models/2650553/anima-edit?modelVersionId=3089149</a> Everything said in that model about usage applies here too (you can even use the same workflow, but use my LoRA instead). But as a recap...<ul><li>This takes an existing image and modifies it, trying to change only the target feature.</li><li>Requires a node to append an image reference latent: <a target="_blank" rel="ugc" href="https://github.com/Anzhc/Anzhc-ComfyUI-Cosmos-Reference">https://github.com/Anzhc/Anzhc-ComfyUI-Cosmos-Reference</a></li><li>Don't add any text to prompt except what you want changed. This one only works with <code>wearing a metal yoke</code></li></ul>The model itself is pretty overfit and always makes the yoke in a specific art style. 😅 If you want to fix the yoke you can mask just the yoke and use the v1.2 LoRA to get it closer to the actual image style.<hr /><h2 id="regular-models">Regular Models</h2>v1.2 - Significantly less style bleed. Slightly less coherence (yoke will look melty or weird more often). Also a smaller sizev1.1 - Slightly less concept and style bleed.A metal yoke that restrains the arms of the person wearing it. Just prompt it with <code>wearing a metal yoke</code>.If the metal is fusing into clothes or isn't well formed, you can prompt <code>metal collar</code> or <code>metal arm cuffs</code> .The only variation is <code>single bar yoke</code> which gets the yoke to have only 1 bar more frequently.Is generally consistent enough from front and back views. struggles with side views.If anyone has any advice on increasing the concept's cohesion so it shows the exact yoke shown in the dataset, please let me know. I used almost entirely 3d renders of the same object from multiple viewpoints and they all had the same physical structure so you think it would generate the same yoke every time.

Metal Yoke

Files with SHA256: f4f81d7c5be5bd58b23caeba24439e4759a135cd1b7b2b3df1a97595a21dd644