Pipkin Pippa of Phase ConnectYT:<a target="_blank" rel="ugc" href="https://www.youtube.com/@PipkinPippa">https://www.youtube.com/@PipkinPippa</a>X: <a target="_blank" rel="ugc" href="https://twitter.com/pipkinpippa">https://x.com/pipkinpippa</a>SDXL 1.0 - Made this lora sometime ago might as well publish it. Worked best with weight 0.6 in my experience. Hair ornaments kinda work.Trained using <a target="_blank" rel="ugc" href="https://civarchive.com/models/199934/flatpiececorexl?modelVersionId=325300">FlatpieceCoreXL</a> + <a target="_blank" rel="ugc" href="https://civarchive.com/models/199948?modelVersionId=224985">FlatpieceXL VAE.</a>10/7/24 - Added a chinchilla version trained using images from <a target="_blank" rel="ugc" href="https://x.com/RageInTheMist/status/1842770838652616912">https://x.com/RageInTheMist/status/1842770838652616912</a><ul><li>Upper body focused, however AnimagineXL has no issues filling in the blanks :)</li><li>Weight of 0.6 recommended.</li></ul>10/16/24 - Added v2 for SDXL 1.0<ul><li>Weight of 0.6 recommended</li></ul><pre><code>base prompt: pippa, rabbit ears, bangs, virtual youtuber, pink hair, pink eyes, ahoge, looking at viewer, hair ornament,</code></pre>12/28/24 - Added a Flux Schnell version using the same dataset as the SDXL v2 model.<ul><li>Weights of 0.75 - 1 seemed to work fine but could vary per checkpoint</li></ul>1/27/25 - Added Flux Dev version<ul><li>Weights of 0.5 - 0.85 recommended but YMMV</li><li>The chinchilla variant tends to appear unprompted adding "tail" to negative prompt should do the trick</li></ul>

Pipkin Pippa (XL/Flux) - Phase Connect

Optimized the dataset for accuracy

PIPPA-XLv2-t4-000003.safetensors

Files with SHA256: b737a087bc03e87fb881e1363cd7748ca1a0e375856bde1ff7d856d0f4d24dec