This LoRA is designed to replicate the photography style of mikkoph (which happens to be me). It focuses on emulating the makeup, lighting techniques, and post-processing effects that I use in my work. The model was trained using images from my portfolio, which I own and hold all rights to.
This LoRA is intended mostly for myself, but also for users interested in generating images in a style similar to my photography, offering a creative tool for experimentation and inspiration.
You can view my photography work at my website: https://michelebalistreri.photo to get a better idea of what to expect.
This is the first LoRA I train and I am pretty sure I didn't do the optimal training, but since it seems to work very well, I decided to post it. It was trained locally on a Strix Halo.
The trigger phrase is "by mikkoph" but it is not really needed. Using "studio shot" seems to be helpful. I suggest using a weight around 1~1.3.
Description
FAQ
Comments (11)
I tested it with several of my fuzz workflows as a support lora. I got some real positive effects for the overall creations to finetune fibre structures. Thx for sharing.
Visited your photo website first - very nice work. I noticed your last blog post is near end 2021 - Probably a coincidence, but that's about when I started to put more and more time into AI and less into the photowork.
Thanks! I basically stopped having time to do photowork in 2022. Got into AI much later, less than a year ago. It is really exciting to be able to realize my ideas once again.
@mikkoph Yes absolutely same for me. I get a lot of hate from my artist and pro photographer friends but I think it's just another creative tool, a very powerful one. Thanks for sharing your LoRA - will try it out.
I rarely leave comments, this LORA and more importantly the style and composition are an order of magnitude above the rest.
thank you! that is really nice to hear
Thank you for making and sharing this LoRA. Any plan for V2? Because this version not only brings the better aesthetic yet also changes character's likeness.
I tried to minimize this by doing the LoRA rank 16 and not using more than 1/2 pictures of each model. Any suggestions of what else I could do? I agree that it does bring some bias towards certain faces but can be mitigated by being specific when prompting
@mikkoph I guess I used low quality character LoRA. I switched to the new training and the likeness is back. Thanks for your quick reply. BTW, both of your Krea 2 LoRAs are amazing.
You are a Mastermind in marketing! Training the lora to place @mikkoph in every image was a brilliant idea to get your name out. But it's unusable for me because of that.
what do you mean? you don't actually need to use the @mikkoph trigger word and the LoRA doesn't put a watermark in the image (as you can see from all generated images here). I still use the trigger word for the images I generate but that's just my choice