Jennifer O'Dell is an American film and television actress.
Update February 12th, 2024:
V2: 500-step TI trained on a new improved dataset of 33 images.
Outdated:
V1:300-step TI trained on a dataset of 11 images.
Description
500-step TI trained on a new improved dataset of 33 images. 10x30 Epochs.
FAQ
Comments (4)
Very good model. Deserves a star rating!
Thank you.
What are your settings for making T.I. now?
Thank you for your review and the question. I still feel like a beginner :D
I don't have any fixed settings. Often a setting with good results for one TI was terrible at the next TI. A lot depends on the dataset.
I currently create between 5 and 15 TIs for a model with varying settings and then select the TI with the best results.
The learning rate is usually '0.05:10, 0.02:20, 0.01:60, 0.004:500, 0.002:1000, 0.001:3000, 0.0005', the settings of Batch size x Gradient accumulation steps / Max steps are between 10x30/300 and 1x1/3000.
Awesome embedding and it's working great! I noticed though that for close-ups, the face starts looking less like her, younger, and more cherubic.
Any thoughts on making a LORA, and maybe a LORA for SDXL and/or Pony SDXL?
I did a few test runs again with checkpoints that I didn't normally use. Unfortunately, the likeness of some of them is not very consistent. The dataset definitely consisted of images from the 'Lost World era' or shortly after.
I only briefly looked into LORAs about a year ago. Looked very awkward and most of the guides were for Colab or other services. I prefer to leave the data on my own computer.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.