I generated 100s of images using several techniques to create a consistent character for a children's book with Midjourney. What did I learn?
It is HARD: Midjourney is not made for consistency and doesn't make our job easy. So if you really want it, be prepared to be persistent and get your hands dirty.
It won't be perfect: Even after generating 100s of images to use dozens, they still won't be perfect. There will be inconsistencies, and the characters will usually be similar but different.
That out of the way, two techniques got the best results for me:
Create a character sheet of images and use it as input
Generate your character in different poses in a single prompt, upscale it, cut the characters, and use all of them as part of your new prompt. Use the same seed for more consistency. Use the template below:
<generated images urls> <character description>, multiple poses and expressions, <image style>, <others> --no outline --seed <seed number>
"Training Midjourney”
Use a "named” prompt to generate your character, upscale your favorite, rate it with heart-shaped eyes, and react with an envelope to get its seed number. Repeat it 8-10 times with the seed number at the end. After that, keep the last seed and slightly change the prompt for poses or emotions. Use the template below:
<character name>, <character description>, <style>, <others> --no outline --ar --seed <seed number>
So, to answer the question. Midjourney is okay to generate characters but could be better. I will use it in a project anyway, but I'm already studying better solutions like Stable Diffusion with Dreambooth.