This is a Stable Cascade Stage C (full) and text encoder fine-tune specifically designed to generate images of the <a target="_blank" rel="ugc" href="https://ultrakill.fandom.com/wiki/Earthmover">1000-THR "EARTHMOVER" </a>from ULTRAKILL. It's mostly a test of how well Cascade can be trained, and judging by the results, the answer is very well. I did several tests using various models (SD 1.5, SD 2.1, SDXL), and this one is by far the best. It handles small details such as the number of eyes, shape and position of the spear, and the environment very well, whereas other models will constantly get even basic things wrong.Note that, however, here, the text encoder is doing most of the heavy lifting; you could generate half-coherent images of the Earthmover even with the base stage C model, however, this fine-tuned one puts the pieces in the right places.Note that these are the raw U-Net and Text encoder checkpoints, so to use them in ComfyUI, you'll need to load them with the UNETLoader and CLIPLoader nodes respectively, with the files being placed in the unet and clip folders, in the models directory. <h1 id="heading-4">Have fun!</h1>

Files with SHA256: f329178b2759abf5aebfa4caa2fc10eb73adacd35836c1721ad52ca86f74b1f6