Recently, there are a lot of partners who want to make animated stories with AI, and we also see a lot of accounts with good data for animated stories with AI.
But unlike doing a 5-10 secondShort Videos, character consistency is one of the most critical issues for mid-length videos.
After all, you can't change a characterization from one scene to the next for the protagonist of a story, can you?
The method of how to keep a consistent character image in AI videos is usually done using MJ or SD, which is relatively complex and not friendly to the little guy.
Today, I'll share a more simple approach that I've recently tested, so that even a white person doing an animated story can simply do consistent characterization.
Primary tool use: i.e. Dream AI. It's friendly to white people.
First, use that dream AI to generate the character three views
We can use Instant Dream AI to directly generate character three-views to create an important character in a story.
Put together a fairly stable set of cue words that people can just use.
Prompt words:Three views, full body, front, side, back. cg anime style character ip* design.Ancient Chinese beauty, dressed in a green hanbok with her hair in a bun.Exquisite features, high detail, best quality, UHD.
(The bolded parts stay fixed, the rest can be modified to suit the character's traits.)
Let's choose one of our favorites and hi-resize it. We can see the character's features from all different angles in the picture.
Second, the use of namely dream reference raw figure to maintain the consistency of the characters
Click on Import Reference Image, select the character's lookalike, and then remember to save it.
Then enter the rest of the cue words, preferably repeating the description of the character's features, and then add the scene and action.
For example: CG anime style, ancient Chinese garden, ancient Chinese beauty, wearing green Chinese dress, hair coiled into a bun, delicate features, woman sitting in a gazebo, stone table in front of her, books, tea cups and teapots.
Choose one of the more appropriate ones and HD-ize it so you can see the results:
You can see that the consistency of the generated figure drawing, and the initially generated three-view figure, is well maintained.
Is not very simple. I believe we can all quickly generate consistent character images afterward, as well as make better animated stories with AI.