txt2img and img2img Workflows

[nic410826]

Hello.
I am pretty new to AI and ComfyUI and would like to get your assistance for the following two 'issues':

1. I would like to formulate a positive prompt in order to create with a txt2img worflow a picture with e.g. three disdinguished persons (i.e. one person with white hairs, one with blond hairs and one with black hairs). I tried different prompts with different formulations, interpunction etc. but some of the characteristics of the three persons on the picture are generated identically (e.g. smiling, red lips etc). How do I formulate a prompt correctly so it will generate a picture according to my specifications?

2. I would like to use an existing photograph which contains certain things I would like to integrate in a newly generated picture. The source picture is only integrated into my target picture if I am reducing the Denoise value to less than 0.5; and in a very bad quality and not positioned correctly at the place where the source picture is showing it. The process does not work at all in a txt2img workflow with the embedding function.; apparently, the embedding function is not carried out at all...

Obviously I am doing something wrong, but I don't know how to correct my error!

Thanks for feedbac!

Regards - Nicola (nic410826)

[ltdrdata]

1. The difference in generation capability varies greatly depending on the model's ability. Models like SD1.5 and SDXL often experience a phenomenon called prompt bleeding and struggle to process long prompts effectively.
   With newer models like SD3.5 and FLUX, prompts can be clearly specified and distinguished through explicit instructions.

2. For this issue, check out a ipadapter.

   • https://github.com/cubiq/ComfyUI_IPAdapter_plus
   • https://www.youtube.com/@latentvision

[nic410826]

I will do so; thank you very much!