Workflow based on InstantID for ComfyUI. It generates random image, detects the face, automatically detect image size and creates mask for inpaint, finally inpainting chosen face on generated image.
The image dimension should only be changed on the Empty Latent Image node, everything else is automatic.
If you get bad results, you need to play with values for masking such as blurring, padding of face bounding box, the usual CFG, steps and sample method and Weight on InstantID node.
Settings in the workflow work fine for me, however Weight on InstantID node should probably be somewhere between 0.70 to 0.80
It's a bit messy and not perfect, but, eh.