ChannelRecipe
Free harness · 3D documentaries

The 3D Documentary Harness

The two rules behind the 3D reconstruction channels blowing up right now: lock the look with Visual DNA, and open every story with Rule 13.

Get the PDF + the next niche lists by email

The harness as a PDF now, then new small channels the Niche Finder catches breaking out. The whole page stays free to read here.

Start here

How to use this harness

Paste the .md harness into your AI's project instructions (a Claude Project, a CLAUDE.md or a custom GPT). Then give it your reference videos and your topic.

  1. Pick 3 or more reference videos from one channel or one niche. Use their best performers, not their latest uploads.
  2. Run Rule 1 on every one of them and fill in the Visual DNA block.
  3. Write your script with Rule 13: the hook first, then the ten beats.
  4. Hand over both together: give your 3D tool or editor the DNA block and the script, and check every shot against the DNA before you render.
The demos from the video

Watch both demos in full

These are the two demos from the video, start to finish. Each one began as a script pasted into ChannelRecipe Studio, which made the narration, every scene, the music and the edit.

B-59: The Submarine That Almost Started WW3Clean 3D Reconstruction recipe
The Killer Mountain Where Dozens Died Before Anyone Reached the TopDeadly Places Documentary recipe
Rule 1

The Visual DNA rule

A trait belongs to the Visual DNA only if it appears in every reference video.
In all of themIt's DNA. Never break it.
In someIt's allowed: use it when the story needs it.
In oneIgnore it. That was a one-off, not the style.

Score the references on five traits, in this order. Each trait has a starter value. Replace it with what you measure.

1Camera: focal length and the slow isometric dolly

Measure
How to measure it
Starter value
Focal length
Do straight lines bend at the edges? Do near objects balloon? Match it to 24, 35, 50 or 85 mm (full-frame equivalent).
35 mm wides, 50 mm mid shots and inserts, never fisheye
Camera angle
How far the camera looks down on the scene
Isometric feel: 30 to 35° down, the camera 1.5 to 2× the subject's height
Dolly speed
Time how long the frame takes to travel one frame-width
2 to 3% of the frame width per second (a full cross takes about 40 s)
Move style
Constant speed, or eased? Does it ever stop?
Constant glide with a 0.5 s ease in and out; never cut mid-acceleration

2Lighting: single-point volumetric light in a dark room

Measure
How to measure it
Starter value
Light sources
Count the shadows each object casts
One key light (a spot or practical lamp), 45° above and to the side
Volume
Can you see the beam in the air?
A visible beam: haze or fog at 10 to 20%
Darkness
How dark the background gets, from 0 (black) to 255 (white)
The background falls close to black (under about 25 of 255); no fill light
Separation
How subjects separate from the dark
A faint rim light on silhouettes

3Mannequins: consistent red and blue shaders

Measure
How to measure it
Starter value
Faces
Are the people detailed or faceless?
Faceless mannequins, no hair, no texture
Colour code
Which side gets which colour, every time
Red = the people in danger
Blue = everyone else. Never swapped.
Material
Glossy, matte or plastic?
One matte shader per colour, the same hex values all video long

4Shot pacing

Measure
How to measure it
Starter value
Hook shots
Median shot length in the first 30 s
1.8 to 2.5 s
Body shots
Median shot length after the hook
3.5 to 5 s
Longest hold
The longest single shot
About 8 s, saved for the money shot
Breathing beats
What happens before each new section
1 to 1.5 s with no narration, music only

53D text call-outs

Measure
How to measure it
Starter value
Position
Where labels sit in the frame
Anchored to the object with a thin leader line, in the upper third, never over the action
Size
Text height as a share of the frame height
About 3.5% of the frame height, uppercase sans-serif
Timing
How long each stays on screen
At least 2 s; one call-out per shot
Entrance
How it appears
A quick fade-and-scale (about 6 frames), no bounce
Tip: watch each reference at half speed with a stopwatch, and screenshot a frame every 2 seconds. Lay the screenshots side by side, and what stays the same jumps out.

Your Visual DNA block

Fill this in from your references and paste it above every script. It's the contract every shot has to keep.

visual_dna:
  references: [video 1, video 2, video 3]
  camera:
    focal_length_mm: { wide: 35, mid: 50 }
    angle: isometric, 30-35 deg down
    dolly_speed: 2-3% of frame width per second, constant, 0.5 s ease
  lighting:
    key: single point, 45 deg high and to the side
    volume: visible beam, haze 10-20%
    background: near black, no fill, faint rim
  mannequins:
    faces: none
    red: "#C8372D"   # the people in danger
    blue: "#2F6FD6"  # everyone else
    material: matte, one shader per colour, never changes
  pacing:
    hook_median_s: 2.0
    body_median_s: 4.0
    longest_hold_s: 8
    breathing_beat_s: 1.2
  callouts:
    anchor: object, thin leader line
    zone: upper third, never over the action
    size: 3.5% of frame height, uppercase sans-serif
    min_on_screen_s: 2
    per_shot: 1
  allowed_not_dna: []   # traits seen in only some references
Check every shot before it renders. If a shot breaks a DNA rule, fix the shot, not the rule.
Rule 13

The investigative hook

Open on three ingredients, in this order, within the first 15 to 20 seconds.

01
Exact coordinates and the vessel's statsThe precise time, the altitude, depth or location, then what it is: its type, its size, how many aboard. Real numbers, spoken exactly.
02
One specific human in a specific seatA name, a role, exactly where they sit. The viewer rides the rest of the video in that seat.
03
The sudden mechanical failureThe moment the machine stops behaving: the part, the sound, the warning light. Don't explain it yet.

The fill-in formula

Exact time and date. Altitude, depth or location above or below place.
Vessel type, size or key stat, number people on board.
In seat or station, full name, role, one human detail.
Then the part fails, in one verb.
Example · B-59, October 1962
October 27th, 1962. One Soviet submarine. Eleven American warships. And one nuclear torpedo. B-59 is trapped off Cuba. Batteries dying. Air thick with carbon dioxide.

Coordinates, vessel stats and a failing machine in under 15 seconds. The human in the seat, Vasily Arkhipov, is held back for the decision point.

Hook rules

Every number is exact and sourced: altitude, time, headcount.
Present tense. It is happening now.
No "in this video we will…". The failure is the promise.
Each sentence gets its own shot, and every shot keeps the DNA.
Rule 13 · The script structure

Where, who, what broke

The hook is only the first beat. The whole script runs on the same three ingredients. These beats fit an 8 to 12 minute video.

1
Cold open
0:00 – 0:20
The three Rule 13 ingredients, on fast hook cuts.
2
Stakes and promise
0:20 – 0:45
One hard number (the odds, the toll, what was at risk) and what we will reconstruct, minute by minute.
3
The machine
0:45 – 2:00
A 3D cutaway of the vessel: the three facts that made it should-have-been-safe. A breathing beat before the next section.
4
The people
2:00 – 3:00
Two to four specific humans: name, seat or station, one trait each. Red or blue by side, all video long.
5
The timeline
3:00 – 7:00
Minute by minute with the clock on screen: normal, then the first anomaly, then the failure, then each decision. Present tense.
6
The decision point
7:00 – 8:00
One person, one choice, in the seat from the hook. Slow down and hold the money shot.
7
The outcome
8:00 – 9:00
What happened, shown plainly.
8
The investigation
9:00 – 10:30
What the recorders, reports or survivors revealed, and what everyone got wrong at the time.
9
What changed
10:30 – 11:30
The rule, design or procedure that exists because of this.
10
The last line
final 15 s
Back to the human in the seat: a quote, a final number, or the one thing they did.

Script rules

One sentence, one shot. Every sentence can be a single 3D scene.
Numbers spoken exactly, and every one sourced.
Never invent dialogue. Quote the record, or say nobody knows.
A breathing beat of 1 to 1.5 s before each new section.
Every call-out names something real: a part, a place, a time, a person.
The last line goes back to the seat from the hook.
The niche list

Proven niches you can break into today

The goal isn't a million-dollar channel. It's a steady $3,000 to $5,000 a month from a niche that has worked for years, with over a billion views between its channels. Every style below is one you can clone in ChannelRecipe today.

Views are for each channel's breakout video. Study the breakouts, not the averages, and run Rule 1 on each channel's top three videos.

Not sure which is yours? Find your money niche in 60 seconds →
01Footage documentary
023D reconstruction
04True crimeall rising channels
06Anime stories
07Kinetic text and motion essay
08Stick figures on detailed backgrounds
11Painting and art-led history
12Hidden-billionaire AI mini-movie
13Dark psychology and stoic essay
Clone any of these styles in ChannelRecipe today: paste three of a channel's videos and Studio learns its look, pacing, music and voice as your own recipe.Start cooking
ChannelRecipe

Skip the days of setup.
Get cooking.

Doing all of this by hand takes days. ChannelRecipe does it for you, on your own computer.

1
Clone a channel's stylePaste your reference videos. Studio measures their cuts, motion, colour, music and sound effects, watches them, and learns the look as a recipe you can reuse. That's Rule 1, done for you.
2
Use the 3D Reconstruction recipePaste a Rule 13 script and press Create video. Studio directs, reviews and renders the 3D scenes.
3
Make it yoursPick your narrator, your palette and your 3D look: a clean white studio or colour film.