A ComfyUI workflow for generating two clearly separated subjects with regional prompt control. The workflow uses separate positive prompt sections for the global image, subject 1, subject 2, interaction, anatomy, scene, and lighting. It also supports global and mask-based regional negative prompts, allowing subject-specific corrections such as preventing gender swaps, duplicate heads, wrong age appearance, merged limbs, and unwanted body-part mixing.The person/subject is always on the white side, depending on which black and white image you download or have created by an AI, and where you use it, you must always describe the white side as a position in the image in the global as well as in the respective person/subject positive prompt.

ComfyUI Regional Two-Subject Prompt Control Workflow

You can now use a switch to choose whether you want to use the model with the model-specific clip and VAE encoder or whether you want to adjust it individually! I have tested SDXL and zImage and they work, but I haven't been able to test the others.I made a small mistake in connecting the model path; the path hidden behind the K Sampler must go to the Lora node for Person 1, then from Person 1's Lora to Person 2, and from Person 2's Lora to the connection behind the K Sampler. Sorry Beginner mistakes! XD

Files with SHA256: 461cd717b7f5d0b99f1d17f302d56e94b957f50b89601379aa1b5ada75ce7655