Last week’s post ended with an honest limitation: I could give each of my D&D heroes a perfect, persistent likeness, but I could not put them in the same picture. Load one character’s adapter and everyone in the frame becomes her. Load both and you get a blend – Jonathan’s armor with Lyra’s hair on a person who is neither. This is the short follow-up where that problem gets solved, with a trick so low-tech it’s almost embarrassing. ...
Everyone who has generated AI images knows the feeling. Each image is fine on its own, but put ten of them on your blog and they look like you hired ten different freelancers who never met each other. The problem isn’t quality anymore. The problem is that the images don’t look like yours. This is the story of how I fixed that with a rented GPU, an evening of training, and a few euros – and what broke along the way. ...
This story was written by the agent-team’s DM agent (Morgan) after our first interactive D&D session. The session was played in real time — me and an AI player agent (Elena, playing Lyra Whisperwind) making decisions, with the DM coordinating combat, narrating the world, and improvising based on our choices. Afterwards, I asked the DM to write up what happened as a narrative. This is what it produced, unedited. The ancient stones jutted from the earth like broken teeth, half-swallowed by centuries of forest growth. Lyra Whisperwind paused at the entrance to the ruins, her amber eyes studying the worn carvings that framed the doorway—symbols that spoke of a time when these halls had been a place of worship, before darkness had claimed them. ...