Milla Jovovich is an Ukrainian-born actress, known for her performance as Alice in Resident Evil. According to the internet, she is also a supermodel, fashion designer, singer and public figure, but I don't know about all that.
The embed has been trained with my usual settings on SD 1.5 and works well on most 1.5 based models.
If you enjoy my creations and want to support me or have a face you'd like to commission, consider supporting me on Ko-Fi or Buy Me A Coffee :)
Description
FAQ
Comments (12)
I can't believe your timing with this. Just yesterday I was wondering how noone did her, and I started on a Lura. Looking forward to trying your version.
Haha yeah that's quite a coincidence, I realised the same yesterday! Let me know when your LORA is up, usually combing a TI + a LORA leads to pretty interesting results :D
@SDKoh It is up https://civitai.com/models/42004 . I never thought of trying the 2 together. Dialing the lora down to 0.3 and they did well together
Great job on this ! One of my favorite ladies.
Thank you!
This may be the best TI out there, not gonna lie. Not only does every gen with it turn out amazing, it looks spot on like Milla Jovovich. Hats off, this one is tremendously good.
Hey thanks man, those are some mighty kind words! I think I got lucky with her as a subject, because she is incredibly well documented and it is very easy to find HQ pics of her :)
@SDKoh Credit where credit is due and no need to be humble about it. You wouldn't be interested in trying a Tessa Fowler TI? I am a bit of a fan :)
@derp123 I could look into that, for sure. Although @JernauGurgeh already has one up and his TI's are very good and consistent. But I could have a go at it as well, it doesn't hurt to have multiple variants of the same face out there!
@SDKoh Absolutely! No problem from my side. :)
I am new in TI training, could u share the training setting, since my training on sd1.5 loss likeness when testing on other base models. How to avoid? Thx for your reply
Hi wgj! So, my general process is summed up here. Since upgrading my GPU, I've started training on a higher resolution, higher batchsize and gradient accumulation. Regardless, most of the steps are still the same :)
It's interesting that you say that training on SD1.5 loses likeless on other 1.5 models. I use SD1.5 (pruned version) to test my TI's. If a TI looks good on 1.5, it tends to work well on other 1.5 models.
I'd recommend trying my workprocess and see if there are any improvements. If there are still issues, you can always ask me here. I reply a bit late sometimes, but please don't let that discourage you from contacting me ^^
@SDKoh Thank you very much for this detailed response. I'll try the guide and report back.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.