OUR PRODUCTION SYSTEM
FireNgine,our productionsystem.
Every piece goes through FireNgine, the production system we built. It plans the piece against the voice, renders it in code and measures every export. Nothing leaves without its proof sheet.
HOW WE WORK
firendev is a video studio built around its own production system.
You bring the script and the voice, or the interview. We agree on one central idea with you before anything is made, and only then ask for the material the piece needs: your product, your logo, your fonts, your footage.
From there, the system handles what should never depend on a tired eye. Every cut, caption and motion cue is timed to the voice. Type and graphics are rendered in code, so they stay sharp in every format and language we deliver. Every export is measured (loudness, safe zones, contrast, caption sync) and the numbers travel with the file; if a check doesn't pass, you hear it from us first.
When something needs to change, we change the plan and render again. Nothing gets patched by hand at the end, so the twelfth version is as clean as the first. And every project leaves notes the next one reads, which is why the work gets sharper piece by piece.
Generative models are one of the tools inside the system. When a piece uses generated footage, a generated character or a synthetic voice, the case study says which shots, and we label them wherever a platform requires it.
Measured.
Every delivery ships with its proof sheet: loudness, safe zones, contrast and caption sync, checked against the platform's spec and kept next to the file.
Consistent across every variant.
One plan drives every hook, cut, format and language. New versions go into the plan and render from it, so the last variant is as sharp as the first.
Visible.
You see the thinking behind the video: the central idea, a hypothesis for each hook, and every note you gave traced to the change it made.
Sharper every piece.
Each project leaves rules, with their evidence, that the next one reads. The system that makes your video has learned from every one before it.
WHAT WE MAKE
What we make.
Eight kinds of video. Each one ships with its proof sheet.
Performance ads for Meta and TikTok
- What you get
- Hook-led ads with three or more openings on one body, cut for 9:16 and 4:5, a 15-second cutdown, and silent loops for product pages.
- What we measure
- Loudness against the platform target, captions inside the feed's safe zone, caption-to-voice sync to the frame, text contrast, the length of every variant.
Short-form social content
- What you get
- Vertical edits for Reels, TikTok and Shorts, one-off or as a steady series, with captions, motion and sound that work muted and with sound on.
- What we measure
- Safe zones per platform, caption reading speed, loudness per platform, faces kept inside the vertical crop.
Product and app launches
- What you get
- A launch film with the interface animated from your live web product, a logo reveal, device mockups and a vertical cutdown.
- What we measure
- UI captured frame-identical run after run, camera moves within 2% of plan, the cursor driven by real actions, logo and motion on their timing tokens.
Explainers
- What you get
- Voice-led explainers built on a motion system: lower thirds, lists, figures, callouts, chapter cards and your own icons.
- What we measure
- Every graphic enters and exits on its timing token, stays on screen long enough to read, and is locked to the voice.
Customer stories and testimonials
- What you get
- From one single-camera interview: a 45 to 60-second testimonial, a 2 to 4-minute story and vertical clips, with caption files, chapters and a music-free version.
- What we measure
- The face kept in frame on every crop, captions timed to the frame (.srt and .vtt), the voice cleaned with its noise floor measured, loudness per platform.
Language and format versions
- What you get
- The same piece in every aspect ratio you need and in English and Spanish, with the on-screen text rewritten in each language; other languages when a native speaker checks them.
- What we measure
- Text reflowed per language and kept inside the safe zone, captions per language timed to the frame, one master per aspect ratio, files named by convention.
Motion and kinetic typography
- What you get
- Type that lands on the word, graphics that move with the brand's timing, and edits cut to the beat of the music.
- What we measure
- Every word on its syllable to the frame, contrast and safe zone for every word, cuts checked against the beat grid.
Generated characters with their own voice
- What you get
- A consistent generated host or character with a fixed voice who presents your product across a series and in more than one language, always declared as generated.
- What we measure
- One character sheet behind every shot, the voice checked against the character's reference voice, every generated shot listed in the case study.
THE SIX STEPS
The voice sets the clock.
Everything hangs on the voice-over. We time every word, then every cut, caption and graphic lands on those times. With no voice, the music sets the clock and cuts land on the beat.
WHAT WE MEASURE HERE: every word's timing, by two methods. KOLDA: 317 words, the two methods a median 27 ms apart (p90 69 ms). Graphite & Stone, with no voice: every hard cut within 0.73 frames of the beat.
One central idea, agreed first.
Before anything is downloaded, generated or cut, we write the piece on one page: what it is in one sentence, what the viewer feels at each moment, the look, the sound, the formats. We agree it with you, and only then does anything get made.
WHAT WE MEASURE HERE: nothing is made before this page exists. KOLDA: the idea, the list of materials and the product, written before the first download.
THE PIECE IN ONE SENTENCE
A runner tells, without talking to camera, how she lives with runner's knee and why there's a steel ball in her freezer; two household images explain it to someone who knows nothing, and it ends honestly: the roller takes the edge off, the exercises fixed the knee.
WHAT THE VIEWER FEELS
One runner and her knee.
Oh, that's runner's knee.
Her own doubt, said before they think it: peas are cold too.
Why steel: the metal bench. And the physical proof: frost on the real product.
Trust, because nothing is promised; an invitation with a condition: if you run.
Then, the materials.
Only after the idea is agreed do we ask for what the piece needs, precisely: product shots, logo, fonts, footage, licensed or generated. Every file that arrives is catalogued with where it came from.
WHAT WE MEASURE HERE: one line of provenance per file.
| id | what | source | where it's used |
|---|---|---|---|
| V04 | physiotherapist moves the patient's knee | licensed footage | section 1 |
| G01 | the product's master frame, 2560×2560 | generated | every view |
| C02 | the frosted ball circling the kneecap, 5 s | generated clip, from still G03b | hook (a), section 4 |
| M01c | the music, from its second 16 | licensed music | the bed |
The plan.
The whole edit is written as data before it renders: every shot with its crop, every graphic with the word it hangs on, every caption with its timing, the voice, the music and each effect. Your notes change the plan and the video renders from it, so the twelfth version is as clean as the first.
WHAT WE MEASURE HERE: every note traced to the field it changed. KOLDA, second round: four notes, four fields, each with its number before and after.
{ "section": "t1", "word": "piled",
"start": 11.93, "end": 12.348,
"f_start": 298, "f_end": 309,
"origin": "forced+onset" }
Rendered in code.
Type, captions and motion graphics are rendered by code, so they stay sharp in every format and language. They move on the same ten timing tokens as this site.
WHAT WE MEASURE HERE: text by code only; every graphic on its timing token (250, 400 and 700 ms in; 200 and 300 ms out).
Measured, then delivered.
Every export is measured on the file we deliver: loudness, true peak, voice over music, text inside the placement's safe zone, contrast, and captions against their words to the frame. We look at every cut frame by frame. The numbers travel with the file; if a check doesn't pass, you hear it from us first.
WHAT WE MEASURE HERE: the proof sheet. KOLDA: every check passed but one, and that one was delivered with its number (ΔE2000 10.42 against 10).
| Check | Target | Measured | Status |
|---|---|---|---|
| Loudness | −14 LUFS ±1 | −14.2 LUFS | passes |
| True peak | −1 dBTP or lower | −4.3 dBTP | passes |
| Voice over music | 10 LU or more | +15.9 LU | passes |
| Text inside Meta's safe zone | all text | all text | passes |
| Worst text contrast | 4.5:1 or more | 4.95:1 | passes |
| Captions in and out, worst | within 1 frame | 0 frames | passes |
| Product the same in every generated shot | ΔE2000 10 or less | 10.42 in one still | didn't pass |
One piece, second by second.
The KOLDA ad with hook (a), as the plan sees it: nine sections timed to the voice, 303 captioned words and 39 graphics, each on its word.
What we don't do.
We don't shoot on location, and we don't supply presenters. When a piece needs people on screen, we use your footage or generated characters, and every case says which is which. We don't buy media or publish for you. And we don't quote views, CTR or ROAS we haven't measured.
Questions
AI is the fuel; FireNgine is the engine, and the engine is where the value is. It is the production system we built: it times every piece to the voice, renders type and motion in code and measures every export before it leaves. AI runs inside it, in parts of the planning and, when a piece calls for it, in generated shots, characters or voices.
FireNgine and the people who direct it make the video. We agree the central idea with you, direct the piece, decide what stays and what goes, and sign off every delivery; the checks on the proof sheet are numbers measured on the delivered file. If a piece uses generated material, you'll know before you approve anything.
No. Generated shots, characters or voices appear only where a piece calls for them; the rest is built and measured in FireNgine.
Yes: no generated images, footage or voices, every shot from your material or licensed footage. FireNgine still plans, renders and measures the piece, with AI in parts of the planning; if that matters for your brand, we'll show you exactly where.
Directed and signed off by a person, built and measured by the system we made. Each case lists what was generated and what wasn't.
Who we are
firendev is a video studio in Pucón, Chile, working with teams anywhere. We direct every piece and sign it off; FireNgine, the production system we built, plans, renders and measures it.
firendev, said fire-n-dev: fire, then develop. In film, to develop is to bring out the image that's already there.
Have a script, a voice-overor a product to launch?
Send a short brief or write to us. We'll come back with questions, then one central idea for the piece before anything is made.
hello@firendev.com