KRISHNA: REBORN is a 3:08 cinematic short — Krishna returns to a rain-soaked neon megacity, and the city does not know what has arrived. Every frame is generated. This page is the part that doesn't show: how the face was pinned down, and what is still wrong with the film.
One frame from every shot, in order. Each is a separate generation — the same Krishna carried across a three-act arc: the awakening, the darkness, the triumph.
The first thirty-five shots were generated with reference images — the usual approach, and the one that quietly does not work. Reference images do not hold a character's identity. They nudge it. Across thirty-five independent generations the nudge accumulates, and the man in shot 30 is not the man in shot 3.
The fix was a registered character, not a better prompt. A proper cinematic character reference sheet was generated first, then registered as a persistent character the generator could be asked for by identity rather than by description. All eighteen Krishna-face shots were re-rendered against it and swapped over the originals.
That re-render hit its own wall: the image model capped mid-batch, and the remaining keyframes came out of a second model. Two image engines inside one continuous sequence is a real consistency risk, and the only check that catches it is looking at the frames side by side.
The film peaks at −0.3 dBFS. That is hot — close enough to the ceiling that some playback chains will not enjoy it. It is inherited from the original mix, and it was deliberately left alone: fixing it means re-processing audio this edit otherwise does not touch, and an unnecessary re-encode is its own kind of damage. It is a known flaw, not an unnoticed one.
Two image models share the sequence. The mid-batch cap means the identity-locked re-render is not the product of a single engine. It survives inspection, but it is not the clean result it would have been.
Three earlier assemblies, kept as they were. All 720p — they predate the 1080p film and have not been re-mastered.
This was made in June. Three days — 13 to 15 June 2026 — and then it sat in a folder for eighty-one days, finished and verified and shown to nobody.
It was never really about the film. I wanted to see how far the tooling would actually go: whether a character could hold identity across thirty-five independent generations, whether a three-act arc could be authored, rendered, scored and cut without a crew, a camera or a location. It was an experiment about possibility, and the film was the by-product.
It went further than I expected it to. Which is the real conclusion sitting underneath all of this: there is a feature-length film in here. Not a proof of concept dressed up as one — the things that were missing in June are the unglamorous ones. Continuity across hundreds of shots instead of thirty-five. Dialogue. A script that earns three acts rather than sketching them. All of that is work, and none of it is a wall.
So it stayed unpublished, because there was no particular reason to publish it. Today there is. Janmashtami — the day that marks Krishna's birth — seemed like the right day to stop sitting on a film about Krishna arriving.