conradlocke

/

krea2-identity-edit

like 545

The LoRA doesn't currently have that feature, but I think it's being worked on.
You should be able to stack another lora with that capability on top of this one though.

The LoRA doesn't currently have that feature, but I think it's being worked on.
You should be able to stack another lora with that capability on top of this one though.

do you know of any lora with that capability?

Just found one and tested it on a few pics: https://civitai.red/models/2759888/zoom?modelVersionId=3105849
The quality isn't the best and doesn't work on all seeds
I just loaded it as a regular LoRA. I didn't use any of the nodes they mentioned.

looking at the examples, it seems that it alters the body.
does it need to be a lora with edit capabilities on top of this or any lora that can generate a close up is ok?

I think you'll be able to use any LoRA that does close-ups as long as you use the trigger word in your prompt, but I haven't tested that myself.
If you want a more reliable way to do this then just cropping the source image will lead to better results. :]

I think you'll be able to use any LoRA that does close-ups as long as you use the trigger word in your prompt, but I haven't tested that myself.
If you want a more reliable way to do this then just cropping the source image will lead to better results. :]

i find that it struggles when setting the output to above 1MP and it is hit or miss with following the prompt, sometimes it doesnt alter anything

Thanks @CeciliaXCIX for the thorough help here. To confirm: dedicated zoom/close-up is not something the model was trained for yet, so it is hit or miss, and cropping the source image is the most reliable path right now, exactly as Cecilia said. It is on my list for a future version. And yes, above roughly 1MP the base model gets less reliable in general, so keeping output around 1 to 1.5MP will also help the prompt-following you mentioned.