Dojo Makes Its Own Intro: Building a Capability Reel From the Inside
Five minutes eighteen seconds. One single-pass master. Every chapter locked by hand. How Dojo produced its own product reel — talk, listen, Lane Assist, try-on, architecture, products, languages, diagrams, kids tutor, and a torii on the last frame.
There is a particular honesty required when you build a tool that claims to help people create.
You can talk about capability forever. You can write white papers. You can ship screenshots with carefully chosen blur. Or you can do the thing that makes the claim legible to a stranger in under six minutes:
Use the product to make the product’s own introduction.
That is what this reel is.
Not a vendor montage stitched by an outside studio that never opened the app. Not a concept film about software. A capability film made inside Dojo — voice, picture, fashion, architecture, product photography, lip-sync, languages, diagrams, a kids tutor, access control, and a finale that refuses to fade out on anything but the mark.
Runtime: 5 minutes 18 seconds.
Encode passes on the ship master: one.
Loudness: −14 LUFS, YouTube-shaped, peaks honest.
Last frame: the torii.
This is the story of how we got there.
Why make our own intro?
Because the first question a serious builder asks is never “what does the brochure say?”
It is: can this thing actually do the work?
Dojo is a Sensei to the green and a grunt for the gray. Mentorship when you are learning. Hands in the dirt when you already know the fix and need another pair of capable hands. Coding is the center of gravity — but modern creation does not stop at a green test suite. Products need stills, motion, voice, packaging, language, safety, and a story people can feel.
So the reel had to prove more than “AI writes functions.”
It had to prove a studio workflow:
- talk to the machine without fighting the keyboard
- listen back without drowning in a wall of text
- run two lanes that actually help each other
- turn a person and a garment into a fashion film
- turn a napkin sketch into a building that looks real enough to walk around
- brand products without losing the label
- speak across languages without chopping consonants
- draw diagrams and hold an equation long enough for a child to learn it
- keep the whole thing safe enough for any age
- end on a mark you would put on a door
If Dojo could not make that film, the homepage had no business claiming it.
The law of the standalone
Early on we made a mistake every ambitious editor makes: we fell in love with cumulative tips.
Section 1 locked. Join it to section 2. Re-encode. Join to section 3. Re-encode. By the time you reach the finale, the brand open has been through twenty-five generations of compression and the first VO has the texture of a photocopy of a photocopy of a good idea.
We killed that workflow on purpose.
Rule: each section’s video/final/ holds only that section’s deliverable. Intermediate cumulatives are previews, not masters. The ship file is assembled once, at the end, from standalones that were approved on their own merits.
That single decision recovered more quality than any codec flag.
The assembler — build-master-reel.py — loads every locked master and title card into one filtergraph:
- normalize geometry to 1920×1080 at 30 fps
- hold each chapter title card for a beat
- dissolve 300 ms into the incoming section (never steal time from the outgoing one)
- pad video or audio when streams disagree by a few dozen milliseconds
- apply loudnorm to −14 LUFS in the same pass
- encode once —
libx264crf 16, AAC 256k,+faststart
No second “normalize later.” No hero pass that ruins the edit. One cut, one truth.
Chapter map — what the reel actually shows
00 · Brand open
A slow morph from the connection hero into the startup welcome. Mic static hunted down and removed until the noise floor dropped from roughly −81 dB to −93 dB. Giants do not live here — they open What is Dojo, where humility belongs.
01 · What is Dojo
Walk on the shoulders of giants. Build with humility. One day, you may walk among them. Then the gem: same giants, new light. The locked intro tip that used to carry brand+chapter as one file was retired; the master now joins the denoised brand standalone to the what-is-dojo standalone without generational debt.
02 · Talk
Keyboard to microphone. The IDE is Dojo’s own — no competitor chrome, no borrowed brand gravity. Zoom is supersampled and integer-centered so the push does not shake. Eve’s VO carries the shift from typing to voice without apology.
02.10 · Listen
Eyes first: a wall of text becomes selective sight. Then ears: natural selection as a listening metaphor, head turn timed, laugh tags cleaned so the take is one continuous human breath — not a stitched joke that forgot its punchline.
02.20 · Lane Assist
Two panels. Solo and Duo. A still of the empty UI, then a phone call where the lanes actually coordinate. The VO is short on purpose:
Dojo is made for coding. Because two lanes work together via Lane Assist. Dojo Solo. And Dojo Duo.
No music bed fighting the phone audio. The call keeps its own sound. That restraint is the feature.
02.30 · More than coding
Dojo-only frame first. Then the capability pills arrive one at a time — multimedia powerhouse without drowning the mark. The VO ends on “and more,” because a honest product never pretends the list is finished.
03 · Fashion try-on
The killer chapter on purpose. Person first, garment second. Preserve face, hands, pose, proportion. Fabric has to fold like fabric. Then motion — a short fashion film, not a slideshow with ambition.
This is where many tools bluff. We did not bluff. The stills and the walk had to survive a freeze-frame test.
04 · Architecture HQ
Napkin sketch in. Photoreal structure out. Then a drone-minded flyover that treats the building like a place, not a texture. Materials have weight. Scale needs a human somewhere in the frame, even if only as a quiet figure.
05–07 · Dojo X products
Combine bar, chips, and can into one commercial still — torii mark only, cursive X, no griffin on packaging. Coffee bag to styled hero. Perfume to orbiting glass. Branding is a constraint, not a sticker you slap on after the beauty pass.
09 · Promo motion
A still becomes an orbit. Commercial grammar without fake luxury fog for its own sake.
09.10 · Lip-sync UGC
Talking-head authenticity is harder than cinematic fog. Lips have to mean the words. Micro-expressions matter. Vertical energy even when the reel itself stays landscape for the site.
12 · Characters
A full-body creature that breathes. Identity held while cloth and weight suggest life. Animation here is restraint — the opposite of thrashing a still until it screams.
13 · Multilingual
Card first: over thirty languages and growing. Then a controlled tour — English, Chinese, Persian, French, Arabic in the reel cut — with onsets detected so consonants are not amputated at the edit. A longer seven-language master lives for blog and deep dives; the reel stays lean enough to breathe.
Heaven is not a metaphor we borrowed lightly. The hill had to feel like a place you could point to while the voice changes under you.
13.10 · Diagrams and equations
State machines. Flow. Pie. A chart that behaves like a chart. Then gravity and light. Then the quadratic formula held long enough that the next chapter can teach it.
Native diagrams are not screenshots of someone else’s slide deck. They are first-class objects in the product — which means the reel can cut to them without shame.
14 · Kids tutor
A girl turns. A tutor named Sarah walks a quadratic with patience. Access control lands the point: safe AI for any age. This chapter is locked as a moral claim, not only a demo. If the product cannot be trusted near a child, it does not deserve a standing ovation for lip-sync.
15 · Finale
FINALE card. Twelve-tile slow-motion montage — tiles stay put while time breathes. Then the torii splash and a download CTA. Linux is fully audible in the end card line because half a platform is not a punchline we accept.
The torii is the last frame of the reel. Not a fade to black. Not a logo sting that apologizes for ending. A gate.
Sound is half the picture
We treated audio like a structural beam, not seasoning.
- Background bed locked and mixed under VO around 12% — present, never royal.
- Brand open denoised so air hiss does not ride every quiet breath.
- Section joins keep outgoing audio intact; incoming fades across the same 300 ms dissolve as picture.
- Multilingual cuts keyed off voice onset, not arbitrary timecodes.
- Final loudness measured, not hoped: about −14.4 LUFS, true peak under −1 dB.
If picture is a claim, loudness is the handshake. A quiet masterpiece that dies on laptop speakers is a draft.
How the files are organized
Everything lives under versioned section folders:
```text
MAJOR.MINOR.PATCH-slug/
before/ after/ video/ audio/ text/
**/final/ ← only this section’s deliverable
```
Majors hold the spine. Minors insert without renumbering the world. Patches hold alternate takes. Remotion and the Python assemblers resolve paths by folder name, not by folklore.
The public ship tree is on GitHub with Git LFS for binaries:
Clone, git lfs pull, and the masters arrive as media — not as 130-byte pointer riddles. Intermediate cumulatives are gitignored on purpose. The canonical master is:
public/sections/15.00.00-final/video/final/DOJO-REEL-MASTER.mp4
From disk to the open web
A master that only lives on one laptop is a diary entry.
We pushed the reel beside the original origin film on the CDN:
- Origin (unchanged):
https://assets.dojoworkspace.io/media/dojo-origin.mp4 - Self-intro:
https://assets.dojoworkspace.io/media/dojo-self-intro.mp4
Then we gave the blog its own production copy of the master — same bytes, stable path under media/production/videos/… — so a future homepage experiment cannot accidentally unpublish this essay’s evidence.
Storage is cheap. Broken links are expensive. We paid the cheap cost.
The marketing homepage now points at the self-intro reel with no poster frame in the way. Press play; meet the product.
What this proves about Dojo
It is fashionable to say AI will “generate content.” That phrase is doing too much work and not enough craft.
This reel required:
- taste — what stays in the cut, what gets locked, what gets deleted even if it was expensive
- systems — section versioning, single-pass assembly, loudness as a gate, LFS as distribution
- evidence — freeze frames, onset detection, measured LUFS, A/V drift checks
- restraint — empty UI before the call, pills one at a time, torii without a circus
- ethics — a kids chapter that ends on access control, not on cuteness alone
Dojo did not replace the human director. It multiplied the director’s reach. Val could lock a chapter, reject a cumulative, demand Linux be audible, and ship a homepage the same day the master measured clean — with a teammate that does not get tired of the twenty-third export.
That is the point of a dojo.
You practice. You correct. You keep the standard. You bow to the craft, not to the hype cycle.
A note on brand
Dojo X packaging uses the torii line mark only and a cursive X. No griffin on products. The animal can live in myth and in motion pictures; the shelf needs a mark a printer can love at twelve millimeters wide.
The splash that closes the reel is the same geometry we would put on a download button. Consistency is not bureaucracy. Consistency is how trust compounds.
If you want to rebuild it
```bash
git lfs install
git clone git@github.com:neekware/dojo-self-intro.git
cd dojo-self-intro
git lfs pull
pnpm install
# optional Remotion studio
pnpm dev
# ship master from locked standalones (single pass + loudnorm)
python3 scripts/check-av-sync.py
python3 scripts/build-master-reel.py
```
You will not get our private takes history — that was never the gift. You will get the locked spine, the assembler, and a README that does not lie about the order of operations.
Closing
We set out to answer a simple charge: show me.
So we showed talk and listen. We showed two lanes that help. We showed a dress that fits the person who was always there. We showed a building rising from pencil. We showed coffee and glass and a dragon that almost breathes. We showed languages without cruelty to consonants. We showed an equation held for a learner. We showed a gate a child can stand behind safely.
Then we stopped talking and put the torii on the last frame.
If you are new, let the reel be a map.
If you are gray-haired in the craft, let it be a proof that the grunt work still gets done.
If you are deciding whether Dojo is real, press play once with good speakers.
Anyone can code.
Sometimes the code is a filtergraph.
Sometimes the commit is a feeling someone can trust.
Welcome to the Dojo.