
Wanted to try and make candid photography style, so I figured an easy way to do that would be forcing the subject's attention somewhere else, like their phone.
Honestly I'm not a big fan of the result, but it does work better than prompts alone. Hopefully enough people find it useful that I can recoup my buzz.
Some things I've noticed with limited testing:
-
it's totally baked, you have to keep LoRA strength low
-
"looking away" doesn't direct their eyes to the phone, so you might want to avoid it
-
eyes often look totally closed, but adding "eyes closed" to negative prompt will cause them to look away from the phone
-
hands are involved, so hands will be fucked up. I don't have any kind of hand fixing in my workflow, but I know it can be done
Example images generated using datassRev3 in comfyui
-
DPM++ 2M Karras
-
20-40 steps, cfg 6
-
adetailer (face fix)
-
Upscaler: 4x_NMKD-Siax_200k
描述:
训练词语: texting
名称: 539346_training_data.zip
大小 (KB): 23680
类型: Training Data
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success
名称: Texting.safetensors
大小 (KB): 223100
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success