For this example, the consistent characters will be Darla and Max from Cats Don't Dance (1997),
both have a distinct style/sizing/coloring etc.
One suggestion is to use an online search engine such as Google and search for a physical description of the character(s) to add to your prompt.
"line drawing, sketch, lines only, Darla Dimples (movie star from Cats Don't Dance) (Darla has fair skin, blonde hair (with her hair styled in ringlets), and baby blue eyes. She wears a light pink dress, a hot pink hair bow, white socks, and pink shoes), standing next to the tall massive colossal strong butler Max. (Max the butler is completely bald and has fair skin, brown eyes and a scar on the right side of his face. He wears a black tuxedo, a white shirt, a black bow tie, white gloves with Mickey Mouse's three slots on hand, and black shoes."
This is the reference image that will be used for the consistent characters with changing expressions. This image was generated by prompting for a line drawing and detailing their appearance. While you can use a reference image from the original source material, that would have influenced their already exaggerated expressions and the coloring. Prompting for a neutral line drawing simplifies the design and allows the AI to focus on the facial expression changes you will prompt in the next step.
Using SeaArt model Infinity (flux) with image to image around .7 strength and then changing which expression you want in the prompt,
"line drawing, sketch, lines only, Darla Dimples and her butler Max, making (happy or sad) face"
This prompt and changing happy or sad face generated both of these images. In the reference image Darla and Max have neutral slight smiles, prompting for exaggerated excessive over reacting etc. will lead to more pronounced and cartoonish expressions.
Closer to .1 strength with image to image will match your original reference image similar to a photocopy, setting the strength higher to .9 will have the ai be more creative and look less like your original reference image.