This LoRA model generates the cute yellow slime that was familiar in Android devices.
Model Details
This is LoRA-C3Lier (with conv2d-3x3). Activation word is blob.
There are epoch-80 model with some flexibility in prompting and epoch-160 model that are somewhat over-fitted. If you use the epoch-160 model, reduce the weights to about 0.7 when applying.
The model was trained on a 192*192 pixel image, it is better to generate with a similar size for icon-like images. Normal images may break down if they are too large.
Total 267 images are used for 89 different images, each with a white, black, or gray background. The prompts are blob and Unicode CLDR Short Name and Keywords (e.g., blob, grinning face, black background, face, grin, grinning face).
It is trained by sd-scripts, network_dim=4, alpha=1, conv_div=4, conv_alpha=1, unet only. See model metadata for details.
Examples
All images are generated with ACertainty, cherrypicked.
