WORKFLOW · Video Generation
Multi-Look Outfit Change Video Workflow
A lightweight workflow for continuous multi-look outfit change videos, one character and one location, seven outfits in a single clip. Originating from the Chinese "empress styling" format, it runs on MiniMax H3 for video with prompts written by GPT-6 Astra, and the lazy version is as simple as swapping the character image and reusing an existing image-to-video prompt. Well suited to fashion and OTTD outfit content.
Here's the thing.
@joshesye (行者AI视频) tried this empress-styling video format a long time ago, and he's upfront about it, the idea came from 小宇 (Chengzilhy). His new clip is the same playbook, just run on H3 this time. He says he's been pushing a lot of video out through H3 lately.
The lazy version, as he describes it, is almost embarrassingly simple. Swap the character image, tweak the image-to-video prompt, done. Back when he first tried it, that prompt was a Seedance one.
His closing note is about where this actually pays off. In his view the format fits OTTD styling content particularly well.
Watch the full clip here, video.twimg.com
At the time of writing, the post sits at 73 likes, 56 bookmarks and 13.3K views.
The quoted post from @Chengzilhy (小宇Chengzi) is where the current version comes from, and it adds the detail that matters.
Last time it was five outfits in one continuous shot. This time he pushed it to seven. One character, one scene, seven looks in a single video.
His stack, GPT-6 Astra for the prompt, MiniMax H3 for the video.
On Astra specifically, he says it genuinely writes prompts more accurately now than before, most visibly on the things this format lives or dies by, the order the outfits change in, the timing of each swap, and the character's movement.
One complaint though. The quota resets every five hours, and a single carefully written complex prompt is basically enough to wipe it out.
His read on who this is for, it suits fashion accounts on Xiaohongshu and Douyin.
At the time of writing, that quoted post sits at 122 likes and 28.1K views. Original thread here, x.com/joshesye/status/2096432790224908773
