<h1 id="prozchoice-rr93ec69c">PROzChoice</h1><h2 id="(5150)-bpnbe2wml">(51/50)</h2><h3 id="note:-bf16-model-is-diffusion-only.-read-this-guide-on-use.-di164wbf1"><a target="_blank" rel="ugc" href="https://civarchive.com/articles/14643/my-model-my-choice-prozchoice">NOTE: BF16 Model is diffusion only. Read this guide on use.</a></h3>FP8 Models are full checkpoints with CLIP &amp; VAE attached.<a target="_blank" rel="ugc" href="https://civarchive.com/models/1050986/fp32-basepony-clip-g-clip-l">FP32 PONYCLIP</a>Currently the FP32 version is exclusive to <a target="_blank" rel="ugc" href="https://tensor.art/models/861372042421738343/Celebrity+PONY-(PDXL-5150)-FP32">TensorArt</a><ul><li>It is a 11B parameter training distilled to allow a perfect blend of SDXL and PONY while using SDXL CLIP</li><li>You can use SDXL named celebrity's or PONY characters</li><li>SDXL LORA's preferred over PONY</li><li>Should work with SDXL control nets as the model was quantized from FP32 CLIP-L and CLIP-G</li><li>The diffusion (UNET) model will work with either SDXL or PONY CLIP-G and CLIP-L</li></ul><hr />Why 51/50 because 51/49 does not sound good, even if it is more accurate.

Files with SHA256: f8fac997e449b00685aed8dc017be730c4dc81eee4a13f560ca679695dbc6ea1