Put your likeness into your favourite painting or re-imagine yourself in a variety of professions

It is based on the nodes from https://github.com/cubiq/ComfyUI_InstantID and  https://github.com/comfyanonymous/ComfyUI_experiments. It requires 12GB VRAM to run this workflow.

Starting from an empty latent we take a square crop of a target pose and apply InstantID. The pair of driving image and pose gives a stable location in latent space, allowing for highly controlled alteration of the clothing and backgrounds.

I could only get (smiling:1.2) & (laughing:1.2) to change the facial expression without altering the whole image. This gives some potential for keyframes to drive a simple animation.

Be careful when using 'spicy' checkpoints. It is too easy for even innocent sounding words to produce images of the chosen subject in the nude.

Tips:

• The prompt should be simple like 'cowboy' or 'firewoman'. If you add things like man or woman dressed as X, then the checkpoint concepts of man or woman will dilute your likeness.
• A high CFG (for Turbo) might be needed to get a good likeness, so the rescale CFG node is used to pull the CFG down to prevent image 'burn-in'.
• You will need to try a variety of checkpoints and images of yourself in until you find a pair that can reasonably  reproduce your likeness.
• Square crop both the pose image and the face for best results. The face needs to be no larger than 640 pixels to be detected.