Too lazy to write the prompt, use an image instead of a prompt.
The vision and the charcater name are brought by Qwen 3.5 with KREA 2 turbo INT8.
Performance 1024x1024: ~11 seconds 5070ti, ~22 sec 5060ti 16g (2nd gen/warm).
The workflow only uses native nodes: no custom nodes, no security risks, and no ComfyUI crashes when you do an update.