For my videos I use Grok. Very good experiences so far, despite all recent complaints about moderation, limits and decreases in quality. I do SFW stuff and pay the SuperGrok sub for 30$ a month.
Now, how I approach the generations
- I created the character sheets that you can see on the pictures. Just asked Grok in quality mode, really. Something like "make me a visual character sheet for [Description]. Generate several views, on neutral white background in relaxed t-pose"
- I tried it also with multiple files with closups, all possible angles etc. What works best so far is really a single image file of 2:3 or 9:16 portrait format, with the front, back and 2/3 iso face closeup. You can do them all in img2img or edit the traditional way
- When I create a scene, I start without the characters first. Then I implant the characters. Then I create video
As you see, my character sheets are really lazy and sloppy made. I don't care. Grok doesn't care too. Someday I make them perfect. Or maybe not.
My prompts usually look like this: "In the tavern, Elara @Image2 and Elias @Image3 sit together at the bar, flirting and laughing. No music. No talking"
I reference the images for the tavern and for the both chars. I add "no music. no talking" because I usually want to add my own music and I don't want the random chatters with artificial voices Grok sometimes creates.
If you need a character in a different outfit, just load the character sheet into img2img and ask to change outfit (pic3). For NPCs I also create character sheets, like for this guard you see on the pic4
To see the results, you're welcome to my youtube channel: https://www.youtube.com/@AlexVkj