This textual inversion file is my first test of this technique. The result seems to me quite convincing but I will understand that it is not up to the fans of this wonderful Korean actress. Sorry! But the face is still pretty and consistent, so usable. I attach some images rendered with different models and it works overall well all the time. I hope to see some prompts with it. Thanks to downloaders :)
This IU aka Lee Ji-Eun embed was trained in SD 1.5, for 3000 Steps using Textual Inversion at a learning rate of 0.5:50,0.05:100,0.005:200,0.001:1000,0.0005:1000,0.0001:2000,0.00005
The Dataset contains 90 images. Embed is 3 vector.
Type the name of the embed in prompt to use in your gens, its "iulje" by default but you can rename it to whatever you please.
Description
1st version
FAQ
Comments (10)
Can you share your training setting details ? I use the same setting with you to train a Chinese female star and it didn't work as well as yours.
So I wanna know what's difference between yours and mine.
Thank you very much
Hi, I follow exactly the process of aitrepreneur in his YT video tutorial : https://www.youtube.com/watch?v=2ityl_dNRNw&t=1052s
Hope it helps !
@aurety Thank you very much for your help
I was not able to reproduce the first image nor the second one. Both prompts also do not include the trigger word. If I add the trigger word, the result still cannot be reproduced. Can you please recheck your promts?
true ! I test directly from my image browser tab ( plugin ) to send to txt2img and I'm not able to reproduce either.. I think it's about the compatibility with emphasis for old prompts, I tried to check the settings and I almost found something quite similar but not close enough.. Sorry !
It's good but I really think it can be improved a lot more. Right now it looks sorta like IU but it doesn't capture the uniqueness of her face. I'd do it myself but I don't have enough vram.
I think it would help the training if all the pictures used are within the same time period, for example 2013-2016 or 2020-2022
Sorry for that, it was my first try on textual inversion ! But the idea to separate the TI based on age is a very nice approach i think ! Hope someone will do the job..
@aurety No need to apologize bro, I think you did very good for a first try - it only needs just a bit of added refinement to capture her face fully. Anyway thanks for sharing it with us.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.