← All posts

One Phone Take, Two Centuries: The Prompts

By TechGuyver · · 6 min read

▶ Read as story

You commented PROMPT, so here is everything. I filmed 15 seconds on my phone at home. Seedance 2.5 in Runway kept my face, voice and every move, and built a stone age village and a Renaissance palace around me. The people in them react to what I do, at the second I do it.

Watch the reel on Instagram

My phone take, the stone age version and the Renaissance version at 2, 6, 10 and 14 seconds

A year and a half ago and now

Same idea, a much better world

In May 2025 I made a video like this with Runway’s Gen-3 Video to Video, in 20 second pieces. It got 490K likes. The worlds looked flat and nothing in them noticed me.

Two things changed since. The model keeps me exactly as I am and changes only the world, and AI now helps write the prompt, so the world plays along with every move. Kids run up when I look out the window. A man jumps out of the bushes the moment I say “whoa”.

The 5 steps

What I did, in order

1. Film one take with moves in it. 15 seconds, phone in your hand. Talk, turn to a window, walk outside, react to something. Every move is a moment the new world can answer.

2. Let AI watch your take before you write a word. Give the video to Claude Code with the prompt below. It pulls one frame per second and a transcript with times, then writes the prompt so each event lands on one of your moves. I could not have timed that by hand.

3. Write the prompt in three parts. The world and the everyday life in it. Then what happens second by second (0-3s, 4-7s and so on). Then: “Keep my face, body, words, voice, movements and the camera move exactly the same. Keep everything else the same.”

4. Keep it one connected place. If you walk from inside to outside, say the outside is the same place you saw through the window. My first Renaissance try named three places with no link, so it jumped to a new set the moment I stepped out.

5. Draft first, upscale the keeper. Seedance 2.5 in Runway with your video as the input, 15 seconds, draft at 480p. Run a few drafts, keep the best one, then upscale it to 4K.

First try jumps to a dark hedge maze after I step out. The fixed prompt keeps the same courtyard and carriage

The prompt that watches your video

Paste it into Claude Code, then drag in your video file

I want to turn a phone video of me into an AI video where I stay exactly the same and only the world around me changes.
Video file: <drag the file here>
New world: <for example a stone age village, a Renaissance palace, a city in 2080>

1. Use ffmpeg to pull one frame per second, with the time printed on each frame, into one contact sheet. Look at it.
2. Transcribe the audio with timestamps (whisper.cpp or any local Whisper).
3. Write me a timeline: for each second, where I am, which way I face, what I say, and every moment the world could react to (a turn, a look, a step through a door, a word like "whoa").
4. Write one prompt for Seedance 2.5 in Runway (my video as the input) in this shape:
   - Start with: "One continuous handheld shot, no cuts, all in one place." Describe the world and the everyday life in it. If I go from inside to outside, say the outside is the same place I saw through the window or door.
   - Then one line per moment, with seconds (0-3s: ..., 4-7s: ...). In each line something in the world reacts to what I do at that second. Build up, and put the biggest event on my last reaction.
   - End with: "Keep my face, body, words, voice, movements and the camera move exactly the same. Keep everything else the same."
5. Check the prompt for jumps: every place I walk into must already be described and nothing appears from nowhere. Then give me the prompt in one code block.

This is what Claude Code looked at for my take: one frame per second, the time on each. With the transcript, it was enough to place every event.

One frame per second of my 15 second phone take, with the time on each frame

My two prompts

Paste into Seedance 2.5, then swap in your own moves

Stone age village

Change the background to a stone age tribe village at daytime, huts of hides and bones, smoke rising, cavemen and cavewomen in furs going about everyday life around me: kids chasing each other, a woman grinding grain on a stone, an elder tending the fire.

0-3s: as I talk, two cavemen behind me stop and stare at me, curious.
4-7s: when I turn and look out the window, outside it a group of hunters drags a huge dead mammoth past, the whole tribe running over to see it.
8-10s: when I turn back and step outside, cavemen in my path jump back and point at me, whispering to each other.
11-13s: as I walk through the bushes, a caveman with a spear crouches in the leaves beside me, tracking me.
14s: on "whoa", he leaps out of the bushes right next to me, shouting, and a spear flies past the camera.

Keep my face, body, words, voice, movements and the camera move exactly the same. Keep everything else the same.

Renaissance palace

One continuous handheld shot, no cuts, all in one place: a candle-lit Renaissance palace ballroom at night with tall glass doors along one wall, opening onto the palace courtyard garden right outside. The courtyard is lined with rose bushes and lemon trees against the palace wall, lit by torches, fireworks in the night sky. Nobles in silk gowns, velvet doublets and lace collars, musicians playing lutes.

0-3s: inside the ballroom, as I talk, the dancers behind me freeze mid-step and stare at my modern clothes, a noblewoman gasps behind her fan.
4-7s: I turn to the glass doors. Through them, in the courtyard, a royal carriage with white horses pulls up, guards with torches run to meet it, fireworks burst overhead.
8s: I turn back, the chandelier above me swinging.
9-10s: I step out through those same glass doors into that same courtyard. The carriage, the guards and the fireworks are still there behind me. Courtiers by the doors drop their wine glasses and back away, pointing.
11-13s: I walk along the palace wall through the rose bushes and lemon trees, the lit ballroom windows still on my right. A masked man in a black cape follows me through the bushes, hand on his sword.
14s: on "whoa", he jumps out of the bushes right next to me and draws his rapier, two guards with halberds rush in from the carriage shouting.

The place never changes: the courtyard I walk into is the one I saw through the doors. Keep my face, body, words, voice, movements and the camera move exactly the same. Keep everything else the same.

Where this goes

You act, AI builds the world

Creativity now sits where the real world you film meets the world you imagine. You still write the story, pick the moments and perform them. AI builds the place and the people, and they play along with you. That mix is a real opening for creators, and I am going to push it much further in a series of tests.

Good to know

  • Draft mode made the whole reel. 480p drafts are quick, so run a few and keep the best before you upscale.
  • The seconds landed. In both of my runs every event arrived within about a second of the move it was written for.

Running a business? Tell me about it and I'll send back what I'd do with AI. Free.