Datasets:

---

malcolmrey

/

various

Like 194

Could you explain this a bit more, is it saying that by prompting details held in the refmod that H3 should just pick them up from the description and generate properly?

'For multi-character scenes, assign mod_1 and mod_2 in the loader and define <Subject 1> and <Subject 2> in your prompt. MiniMax-H3 will automatically map the identity latents without token collisions.'

Also if you are creating audio refmods, how would you prompt to ensure a particular character uses a particular voice?

Thanks for releasing these!