The Last Transmission — A 3 Minute AI Generated Sci-Fi Short Film About an Astronaut Who Discovers Her Mission Was Never What She Was Told
Built by TectonicStream ·
About this build
A fully AI generated 3 minute cinematic sci-fi short film built across 6 weeks using Midjourney, Kling, ElevenLabs, Suno and Runway — featuring a solo astronaut stranded on a research station who intercepts a transmission revealing her mission director has been lying to her from the start.
What was built
The story: Commander Yael Ashford has been alone on Research Station Vega-7 for 14 months. Her mission — collect atmospheric data from a dying planet below and transmit it back to Earth. On day 412 she intercepts a partial transmission not meant for her. What she hears in those 43 seconds changes everything she thought she knew about why she was sent there. The 3 minute film covers the 11 minutes between intercepting the transmission and making the decision that follows. Why this story: I wanted a story that could work in 3 minutes without feeling rushed. That meant one character, one location, one decision. The research station setting also gave me visual control — a consistent interior environment is much easier to keep visually coherent across AI tools than multiple locations. The emotional arc had to be simple — confusion, realisation, decision. Three acts in 3 minutes. The production process: Visual world building — Midjourney The research station needed to feel real and lived-in rather than like a clean concept art render. Spent 8 days establishing the visual language before generating a single frame for the film itself. Core style prompt used throughout: claustrophobic deep space research station interior, worn industrial surfaces, amber emergency lighting mixing with cold blue monitor glow, condensation on windows, 2040s near-future aesthetic not science fiction fantasy, hand-held documentary camera feel, slight lens aberration, photorealistic, 16mm grain The protagonist Yael took 4 days to lock in. The brief was a 38 year old woman who looks like someone who has been alone for 14 months — tired around the eyes, hair practical not styled, competent hands, the kind of face that looks like it has made hard decisions before. Once I had 5 strong reference images locked in I used them as image prompts for every subsequent Yael generation throughout the film. Motion — Kling and Runway Used Kling for all character motion shots — close-ups of Yael's face and hands, her reactions, the moment she hears the transmission. Kling handled the subtle facial expressions better than any other tool tested at this scale. Used Runway for environment shots — the planet surface through the window, the exterior station shot that opens the film, the slow push-in on the transmission console. Runway's camera motion controls gave more cinematic movement on these wider shots. The most technically challenging shot was Yael's hands pressing the playback button on the console. Hands are still the hardest thing to get right in AI video — the tendency toward extra fingers or melting joints is well documented. Solved this by generating the hand shot as a static Midjourney image first with explicit hand anatomy instructions, then using Runway to add minimal motion — a slight lean forward and a single button press — rather than asking Kling to generate hand motion from scratch. Voice — ElevenLabs Yael needed two voices — her spoken dialogue and her internal monologue narration which overlays several silent sections. They needed to feel like the same person in two different emotional states. Built a custom voice using ElevenLabs voice design. The spoken dialogue voice was calibrated for exhaustion — low energy, slightly hoarse, the voice of someone who has not had a proper conversation in months. The internal monologue voice used higher stability and slightly faster pacing to contrast the urgency of her internal processing against the flatness of her external speech. The transmission voice was a deliberate contrast — clinical, well-rested, the voice of someone who has never spent a day in space and is making decisions about someone who has. Generated this as a separate ElevenLabs voice design and then processed it through an additional layer to add radio compression artefacts. Soundtrack — Suno The score needed to do something specific — it needed to feel like it belonged to a 1970s NASA documentary for the first 90 seconds and then gradually become something much more sinister without the listener being able to identify exactly when the shift happened. Prompt that produced the closest result: documentary space exploration score in the style of late 1970s NASA archival footage, warm analogue synths, hopeful but slightly melancholy, solo oboe over sustained string pads, gradually introduce discordant undertones under the main theme, never fully resolve into horror but make the listener increasingly uneasy, no percussion until the 2 minute mark Took 14 Suno generations to get the transition right. The earlier generations were too obvious about the shift — they went from hopeful to ominous too cleanly. The version that worked sounds like the same piece of music the whole way through but leaves you feeling different at the end than you did at the start. The edit: Edited entirely in a standard video editor cutting between Kling and Runway clips. The discipline was keeping every shot under 4 seconds. Longer than 4 seconds and the AI inconsistencies have time to become obvious. Under 4 seconds the eye accepts the visual before it can interrogate it. What I would do differently: The exterior station shot at 0:14 does not match the interior lighting temperature. This is the most obvious tell in the film — the exterior reads as a Runway shot in a way that breaks from the Midjourney world. In the next version I would grade the exterior footage to match the amber-blue interior colour temperature before cutting to it. The transmission audio artefacts were too clean. Real radio compression sounds messier than what ElevenLabs processed audio sounds like. Would add a custom audio degradation layer in post to make it feel more genuinely intercepted. What I learned: The emotional resonance came almost entirely from the voice performance and the score — not the visuals. The visuals needed to be good enough to not break the illusion but the feeling came from sound. This changes how I would approach the next project — I would spend more time on the voice design and sound design and be less precious about visual perfection. Three people who watched an early cut described feeling genuinely tense during the transmission scene. One said they caught themselves holding their breath. That was the moment I felt like the experiment had worked. Prompt used: Core Midjourney style: claustrophobic deep space research station interior, worn industrial surfaces, amber emergency lighting mixing with cold blue monitor glow, condensation on windows, 2040s near-future aesthetic, hand-held documentary camera feel, slight lens aberration, photorealistic, 16mm grain Yael character: 38 year old female astronaut, tired around eyes, practical short hair, competent hands, worn but functional mission suit, the face of someone who has been alone for 14 months, photorealistic, warm skin undertones against cold blue monitor light Suno score: documentary space exploration score late 1970s NASA archival footage, warm analogue synths, hopeful but slightly melancholy, gradually introduce discordant undertones, never fully resolve into horror, no percussion until 2 minute mark Behind the scenes: Total Midjourney generations — 280 plus. Total Kling clips generated — 34. Total Runway clips generated — 18. Total Suno generations — 14. Total ElevenLabs voice iterations — 22. Total working time across 6 weeks — approximately 55 hours. Longest single session — 9 hours on the Yael character design.