Type | |
Stats | 1,246 1,560 |
Reviews | (260) |
Published | Mar 25, 2023 |
Base Model | |
Training | Steps: 1,700 Epochs: 5 |
Trigger Words | mntydge doge |
Hash | AutoV2 A8CD183CA3 |
A LoRa of Kabosu aka Doge from the meme, link to her owner on Twitter
Based in Realistic Vision 1.4 but has success across most models based in SD 1.5
Trained on almost 40 high quality doge memes and art I've done as Minty Doge on Twitter.
You don't need an invocation tag as it tends to apply the style to anything with a head and looks ok just invoking "doge" but the trained tag is mntydge
Best results with inpainting and/or scaling with a Lora weight 0.8-1 and CFG 5-7
Though lower weighting is needed to move it away from realism.
I'm aware of lower network dimension but it's push and pull for me until I get more practice training Loras so this is at 128 / 128
Realistic example images are in Realistic Vision 1.4, non realism ones are with Anythingv3