Tonight I finished the pipeline that generates short-form video for Tony, my AI gym bro character, and posts it to TikTok, Instagram, and YouTube Shorts. One command builds the clip, burns the captions, writes the caption and hashtags for all three platforms, and drops everything in a folder ready to review. That part wasn't the hard part. Generating video is basically solved now. The hard part is making an AI keep coming up with something different every single day. Left alone, the model converges. Ask it for "a gym joke" a hundred times and you get the same five jokes wearing different clothes: leg day, protein, skipping the gym. Same problem hits the visuals. Ask for "an animal doing something funny" enough times and you get the same three animals in the same three poses. The fix isn't a better prompt. It's forcing a structural constraint every single time, then keeping a ledger of what's already been used so I never repeat a mechanism, a joke setup, or a scene two days running. That's the actual unsolved problem I'm working right now, not the video generation. I've got the pipeline running end to end. What I don't have yet is proof the format itself holds up past week one. I'll drop the actual clips here once I've got a few days posted so you can judge for yourself instead of taking my word for it. Anyone else fighting the "AI keeps giving me the same output" problem? What's worked for you to force real variety instead of a better prompt? Here's the results for tonight. I'll be setting this up to run a few times a day. Right now I must do the posting itself. I've automated that too it's just not part of the pipeline yet: https://www.youtube.com/shorts/yw6SfIZOm9Q