Emma Watson
Trained under a SDXL model using clear HD images
Trigger word = Emma101 V2 emzwats101 V3 emmawatsonv3 V4 emmawatsonv3
positive words = <lora:Emma_Watson_SDXL_101-000010:1> 1girl Emma101 outside light smile looking at viewer [ or when using v2 emzwats101 or V3 emmawatsonv3 emmswatons101vepic
Note the weight on V3 is about 0.8
Negative words to use = ugly, deformed, airbrushed, photoshop, rendered, (multiple people), child , young
please leave a like and share
Description
FAQ
Comments (4)
i like the this new one better well done
thanks
Thank you again for making another beautiful SDXL LoRA. This is much better than v1. However, half the angles didn't seem to reflect the likeness. But when I mix both together such as "<lora:Emma_Watson_V2_SDXL_101:0.8> emzwats101, <lora:Emma_Watson_SDXL_101-000010:0.8> Emma101", the likeness is higher and nearly all angles work/are consistent. I'm very pleased with this and it's now the best SDXL Watson. I had trials with different weights at the same seed and found if both are at 1.0+ or if I mixed at different weights it was wonky. I think more trials could be done to determine the absolute best. I'm sure it's somewhere between 0.72 and 0.89.
"Thank you for the feedback. I intended to include more full body shots, but I feel they often dominate the focus. I might revisit her in a week or so to create a v3, aiming to perfect it by learning from my mistakes and retaining the successful elements. I won't release a new version unless it offers improvements over this one."