← All posts

My Phone Is Still a Portal: 6 Prompts Later

By TechGuyver · · 7 min read

▶ Read as story

You commented PROMPT, so here is everything. In May 2025 my phone was a portal: one walk from my room out to the balcony, turned into the stone age and Byzantium, 17.6M views. Today I made it again with Seedance 2.5. This time I didn’t write the prompt. Opus 5.5 did, from my notes, over 6 versions.

The two videos

Top is my real take, middle is today (Seedance 2.5), bottom is May 2025 (Runway Gen-3)

Constantinople, 941 AD, the day Greek fire burned the Rus fleet

A stone age cave house, the day a rival tribe attacks

What changed, version by version

Same take, same second (12.5 s, the reveal), six prompts

Byzantium, versions 1 to 6 at the same second

Stone age, versions 1 to 6 at the same second

My first prompt was the one that worked for my caveman remake: who I become, what the room becomes, keep my motion. About 90 words. It gave me a pretty, empty palace and a calm cave. Most AI models give you the bland version until you tell them what matters. So after each try I told Opus 5.5 what was wrong, and it wrote the next prompt.

  1. Just the era. “Turn this into the Byzantine Empire”. Nice room, nothing happens.
  2. Timed beats. “At 7s he steps out, the big reveal”. Name the second and it lands there. In the stone age, long scene descriptions swapped me for a different man.
  3. Keep my face, bright from the first frame. “His own face stays clearly his, same eyes, nose and expressions.” “At 14s he turns around and talks to the camera.” “Bright and sunlit from the first frame”, or you get a candle-lit room.
  4. Every real thing stays in place. List what is in your shot and what each thing becomes, in the same spot and size: the wooden door becomes a carved cedar door, the TV becomes a gold mosaic icon, the balcony door becomes a marble archway. Runway used to have a structure slider for this. Now the prompt does it.
  5. A day in history. Not a room, a moment: people doing real things, live sound, the big event on the reveal.
  6. Action. Things fly and fall, people pass right around me, a battle when I step outside. The doorway I walk through is an open gap, or it puts a door there and I walk into it.

Version 6 is the one I posted. About 340 words, and I wrote almost none of them.

The prompt that writes your prompt

Paste into Claude or ChatGPT with your video’s details

I'm making an AI video to video clip with Seedance 2.5 in Runway, edit mode. My input is a phone video of me. Write the prompt for it.

My shot, in order: [what happens second by second, e.g. 0-6s I walk through my room talking, 7s I step out onto the balcony, 14s I turn around].
What is in the frame: [every real thing and where it is, e.g. big wooden door on the right, TV on the back wall, bed, balcony door, railing, the view].
The world I want: [an era or a place] on [a real day or event, if there is one].

Rules for the prompt:
- Overlay the new world on the real one: every wall, opening and object stays where it is, same size, and you say what each one becomes.
- I change too: clothes, hair, body. Nothing modern stays on me. My own face stays clearly mine, same eyes, nose, expressions and mouth movements.
- Time the beats by second, with the big reveal at the moment I step into a new space.
- Make it a moment, not a room: people doing real things around me, things moving, live sound.
- Bright and sunlit from the very first frame.
- Never describe my gestures. End with: "Keep his exact motion, gestures, timing and the same camera."
- Never put a child in an action or attack scene (Runway blocks it).

After I run it, I'll tell you what's wrong and you write the next version.

Then do what I did: run it, look, tell it what’s wrong in plain words (“too dark”, “that’s not my face”, “nothing happens”), and run the next one.

My two final prompts

Version 6 of each, as I ran them

Constantinople, 941 AD

Restyle this exact video as a wing of the Great Palace of Constantinople in 941 AD, the day the Rus fleet attacks the city and the Byzantines answer with Greek fire, bright and sunlit from the first frame. Overlay the new world on the real one: every wall, opening, object and the floor stays exactly where it is, same shape and size in frame, only what it is made of changes: the big wooden door on the right becomes a carved cedar door with gold inlay, the white walls become white marble with gold mosaics, the TV becomes a gold mosaic icon, the bed becomes a couch with silk cushions, the balcony door becomes an open marble archway, the railing becomes a marble balustrade.
0-6s, inside, everything is happening right around him: servants with trays of food and wine weave around him, guards in armor run past his shoulder, a courtier ducks under his arm, dust falls from the ceiling with each distant blast, a silver goblet drops and bounces by his feet, a chandelier swings above him.
At 7s he steps out through the archway, the big reveal: the Bosphorus below is a sea battle, Byzantine warships spray roaring Greek fire across the water, enemy ships burn, flaming arrows and catapult stones arc over the city walls, Hagia Sophia's dome in the sun behind it all.
8-15s: on the terrace, archers run past him to the balustrade and loose arrows at the ships.
The man himself: his own face stays clearly his, same eyes, nose and expressions and every mouth movement as he talks. He becomes a Byzantine nobleman in a deep blue silk tunic and a gold-embroidered cloak. Nothing modern stays on him.
He moves calmly through all of it, as if he knows exactly what happens next; everything happens around his real path, never changing his motion. At 14s he turns around and talks to the camera. Keep his exact motion, gestures, timing and the same camera. Cinematic, intense, photoreal, bright natural daylight, rich color, loud live sound of the scene.

Stone age

Restyle this exact video as a stone age cave house on the day a rival tribe attacks. Overlay the new world on the real one: every wall, opening, object and the floor stays exactly where it is, same shape and size in frame, only what it is made of changes: the big wooden door on the right becomes a wall of lashed logs and hides, the white walls become warm sandstone and mud with charcoal paintings, the TV becomes a painting of a mammoth, the bed becomes furs on a stone shelf. The balcony doorway stays an open gap in the rock with nothing in it; he walks through open air.
0-6s, inside, everything is happening right around him: tribe members rush past him grabbing spears and torches, a woman carrying a basket of firewood hurries past his shoulder, a man ducks under his arm, burning embers and ash fly through the air, a clay pot falls and smashes beside him, a spear flies past behind him and sticks in the wall.
At 7s he steps out through the opening, the big reveal: the valley below is a battle, warriors with torches and spears charge up the slope, a herd of woolly mammoths stampedes through the smoke, fires burning in the grass, arrows flying overhead.
8-15s: on the rock ledge, hunters run past him on both sides into the fight.
The man himself: his own face stays clearly his, same eyes, nose and expressions and every mouth movement as he talks. He becomes a buff caveman with long wild hair, a thick beard, a bare chest and an animal-hide loincloth. Nothing modern stays on him.
He moves calmly through all of it, as if he knows exactly what happens next; everything happens around his real path, never changing his motion. At 14s he turns around and talks to the camera. Keep his exact motion, gestures, timing and the same camera. Cinematic, intense, photoreal, bright natural daylight, rich color, loud live sound of the scene.

How to run it

  1. Film one walk. Your phone, one take, 16 seconds or less, moving from one space into another (a room to a balcony, a hallway to the street). The step through the doorway is your reveal.
  2. Upload it to Runway. Seedance 2.5, edit mode, your video as the input. New to Runway? Sign up with my Runway referral link.
  3. Pick 720p. Each of my runs was about 16 seconds at 720p. Look at the credit price on the screen before you press generate.
  4. Run, look, tell your AI what’s wrong, run again. That loop is the whole method.

Good to know

  • Related. The caveman and Florence remake of my most viral reel is in Video to Video, 22 Months Later.
  • The model is not the bottleneck anymore. Seedance shoots it, the AI writes it, you direct. Knowing what you want, and saying what’s wrong, is the skill.

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