If you like what I do please consider supporting me on Patreon and contributing your ideas to my future projects!
Frank Frazetta art style lora for SDXL 1.0. Frank Frazetta is a legendary fantasy painter and now you can generate art similar to his signature style within your own Stable Diffusion workflow. This model was trained on 1024 x 1024 and is able to produce higher details than my 1.5 model.
I use this at a weight of around 0.75 but sometimes I'll have to go down to 0.4 depending on the prompt. I used clip skip 1 for all of the images you see uploaded to this page.
Have fun playing around with this remarkable tech and if you like it be sure to review and upload all of your amazing works of art!
Invite to my public discord: https://discord.gg/UFkPTfTDWs
Description
FAQ
Comments (13)
This looks great!
Question, I'm having lora trainings with Kohya_ss for SDXL fail out far sooner than those I did for 1.5, or I'm running out of mem. Do you find you had to change settings, like use less images, or change batch or repeats? Open to sharing config files/settings overview?
Great job. You really captured Frazetta's essence. Too bad SD can't do hands and gets way too "creative" sometimes!
Thanks I'm glad to hear that, for hands photoshop beta has been a godsend, it's a must have for making art with stable diffusion
@getphat You're using generative fill on the AI hands? Does it match the style of the image well? I figured with Adobe's AI, it would make everything photorealistic.
@SnowSultan If a gen fill hand doesnt match the art style enough you can go over it with Harmonization neural filter and that usually does the trick
Throw Boris Vallejo in the mix and it's even more epic I suspect! Looks great.
I was a bit surprised the art style was trained using only 1,500 steps and 1 epoch. How many training images did you use? Regularization images? Based on the LoRA file size I assume your network rank/dim was 128 to 192 and alpha network was 2 to 8 range? What was your learning rate and LR scheduler?
Also, I have a 4070 Ti 12GB and I'm able to do a batch size of 4 but I have to enable a bunch of GPU memory efficiency settings including Gradient Checkpointing.
I have a hand full of different training methods I use depending on the subject/style, catch me on discord or patreon if you ever want to chat
I'd love to see a Larry Elmore one too!
Me too
+1
Thank you for your time and effort I love your art style..........
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.