He’s been trying to get believable performances out of Minimax.
His post didn’t get a lot of attention so I wanted to post showing just how much of an impact it has
You have to prompt to be consistent with the actor you’re trying to replicate, then here’s what you add to the prompt to guide minimax.
<Subject 1> is the [character] whose entire appearance comes exclusively from <Picture 1>: [appearance]. [She/He] speaks every line as an impression of the speaker in <Video 1>.
<Subject 2> is the [character] whose entire appearance comes exclusively from <Picture 2>: [appearance]. [She/He] speaks every line as an impression of the speaker in <Video 1>.
<Video 1> is the delivery reference for both characters. They copy the speaker’s cadence, pauses, pitch changes, emphasis, tone and facial manner. Ignore the speaker’s appearance, voice timbre, words, setting and framing, and omit all text and watermarks.
<Video 1> (delivery style): fully_preserved – the speaker’s cadence, pauses, pitch changes, emphasis patterns, tone and facial manner are applied to both characters, while the speaker’s appearance, voice timbre, words, setting and on-screen text are excluded.
Both characters deliver every line as an impression of the speaker in <Video 1>: the same pauses, pitch changes, emphasis and facial manner.
