Daria and Daria2 are a bit of an experiment in how different body types affect training. Daria is skinny like a heroin-chic supermodel, whereas Daria2 is more in the Playboy model category, but they do have quite similar faces.
Generated on 147 images created solely by prompts and bodysliders, 15 vectors, learning rate 0.004, 15 steps pr image/embedding and 1650 steps in total. Responds well to all manner of checkpoints.
Description
FAQ
Comments (1)
If anyone is receiving byteorder and serialization_id errors for tvange365's TI files when starting Automatic1111 and the files are absent in A1111, please read...
To correct this, open the affected .pt file with WinRAR, 7-zip, or equivalent. Delete the byteorder file and the .data folder, then save the .pt file (some archivers save automatically).
Problem solved and A1111 will load and use the embeddings as expected.
Note that this is NOT tvange365's fault. This is a known issue with how A1111 creates the pickletensor files and how they conflict with the safety checker on certain A1111 versions.
https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/3878
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.