💥 BIG breakthrough in AI video today: Minimax by Hailuo AI (not affiliated but I use it on Photo AI) launched a "subject reference image" feature The problem with AI videos right now is that you can generate great videos with just a prompt, but they won't have your AI clone in them, they'll just have a random AI person in them So you can't make videos with YOU in them for example So my hack for that was, generate an AI photo first, then use that as the first frame (img2video) and it then EXTENDS that frame, but you don't get as great videos as just free form prompting videos, because AI models are not as good at extending frames as they are at just free form generating videos without any constraints of a first frame Another problem we all had with img2video is that your character does not remain consistent, so you start with your AI clone's face, but then it slowly evolves into someone else By adding an AI photo or real photo as a "subject reference image", it will use that to keep the character (e.g. the face) consistent throughout the video Which now opens it up to be actually useful for creating real videos, films etc. Here I tried it: Minimax + Subject reference image (from Photo AI) + MMAudio (AI generated audio) + "a woman is walking through a busy Tokyo street at night" as free form AI video prompt Amazing result I think: