Lora of comedian Sarah Silverman.
V2.0
Trained with 2379 steps on 405 images (768x768).
Trigger: S4r4hS1lv3rm4n
Useful prompts: 1girl, black hair, brown eyes, lips, long hair, jewelry, earrings, breasts, cleavage, smile, teeth, bare shoulders, parted lips, grin, collarbone, makeup, nose, necklace, open mouth, short hair, lipstick, pearl necklace, ponytail, red lips, closed eyes, closed mouth, one eye closed, brown hair, hair bun, tongue, tongue out, messy hair
The dataset has been processed using a new workflow. Read more about the dataset preparation here: Faster image editing (Photoshop workflow)
V1.0
Trained on 159 images (768x768, 3 epochs). Trigger should not be needed if <lora:S4r4hS1lv3rm4n_V1:x> (x = weight) is included in the prompt.
Useful prompts: black hair, jewelry, earrings, brown eyes, looking at viewer, long hair, smile, teeth, dress, black dress
Due to the nature of the AI, it might be possible that the generated images contains nudity and/or NSFW content. Please use appropriate negative prompts to counteract this.
Description
FAQ
Comments (1)
This looks undertrained. From the example images you posted I couldn't even recognize her (the embedding doesn’t look convincing either). Her face isn’t so complex that you should need to combine it with an embedding to get a good likeness.
When trying it out, it isn’t that bad if you increase the weight. Might have needed another epoch to come closer.
With such a high number of training images you might want to increase the Network Rank (Dimension) and the Network Alpha and see if that improves the results.