V1Automatic video editing · auto captions · English, Malayalam
EDITSON is an automatic video editing engine. It takes raw footage and returns a finished, on-brand video: smart cuts, auto captions, B-roll, graphics and mixed audio. It runs on our own NVIDIA GPU hardware, so long 4K footage never leaves the building.
V1What it does
One engine handles the repetitive parts of editing: picking takes, cutting dead air, captioning, cutting to B-roll and mixing sound. A person reviews the result before it ships.
Multi-camera podcasts are synced, cut to the speaker, cleaned of microphone bleed and finished with a cold-open hook, chapters and titles. A 71-minute 4K two-camera session became two finished episodes.
See it in action ›Word-by-word captions land on the exact moment each word is spoken. A person proofs the text. We have shipped English and Malayalam, including speech that mixes the two, and a dedicated model covers 22 Indian languages.
See it in action ›Vertical 9:16 edits with captions, spec cards, stamps, B-roll cutaways and a music bed that ducks under the voice, mastered to the platform loudness target.
See it in action ›Fifteen-minute films with chapter labels, on-screen overlays, a thumbnail and a post kit with titles, description and chapters.
See it in action ›Studios and software teams use EDITSON under their own brand, either as a production service or integrated behind a podcast or editing product. Start with a one-episode pilot.
See it in action ›A1The stack
CUDA runs the speech models and matting. NVENC encodes the final cut. We work with NVIDIA and AMD, and our content is built around the hardware both make.
CUDA pipelines and NVENC rendering on GeForce RTX.
Ryzen and Radeon builds are a core part of our content.
A flagship-class CUDA GPU with 24 GB of video memory for long 4K sessions, on site.
GPU speech recognition, matting and hardware video encoding.
Partnership wording and logos are subject to each partner's brand guidelines.
T1Algorithmic editing
An edit starts with a question: which pauses are dead air and which are timing? We measure the audio and answer it with numbers. Move the slider to see how one rule reshapes a take.
Illustrative signal, generated in your browser. The pipeline applies the same rule to real voice tracks, with extra checks so words are never clipped.
Energy, pitch range and pace pick the best take of every line.
Face and lip tracking keep the speaker framed and sync extra cameras.
Word-level timing puts every caption on the exact moment it is spoken.
V1The pipeline
Each stage is its own module, so a partner can adopt one stage or the whole line. Watch the footage move through.
Sync every camera and audio file to one timeline.
Know what was said, who said it, and where the face is.
Choose the best take of every line and plan the edit.
Turn the plan into pictures.
Make it sound as good as it looks.
Check the cut before anyone else sees it.
T1Visual styles
Style is data: fonts, palette, caption behaviour and how graphics enter. We load a new look per brand without touching the cutting logic.
High-energy captions that pop in word by word, with spec cards and stamps landing on the exact word.
A cooler build-reel look: ice-white type with cyan and violet neon cards, full-screen B-roll and stacked shots.
Small, readable English subtitles a phrase at a time, with a product end card. Built for a spoken-Malayalam audience.
Client and partner marks stay in the safe zone for the whole reel; white pill subtitles never collide with them.
When the footage is already graded and beautiful, we add nothing: clean cuts, clean audio, a strong cold open.
A2Languages
We have shipped captions and subtitles in English and Malayalam, including speech that mixes the two. A dedicated Indian-language speech model covers 22 constitutionally recognised languages. Each language is benchmarked on your footage in a pilot before we commit to it.
Shipped across 20 finished pieces, each with a human proofing pass and English subtitles.
Hindi, Tamil, Telugu, Kannada, Bengali, Marathi, Gujarati, Punjabi and the rest of the 22 official languages, plus English, Bhojpuri and Bhili.
We test your language on a real episode and measure accuracy before delivery. Names and terms get a human check.
No match. Ask us: new languages are added per project.
Support means automatic transcription plus subtitles with a human proofing pass. Accuracy differs by language, and we measure it per language before delivery.
V2The work
Excerpts from finished videos for real shoots. Everyone shown has given consent for pitch use. Drag or flick the row, use the arrows, or tap a clip for sound.
A2Build log
Dates come from our project files and are rounded to the day. Each entry is a finished deliverable, not a prototype.
Captions, cards, grade, music and sound rules become a style guide the pipeline follows on every video.
Word-by-word captions, spec cards, cutouts over B-roll and a ducked music bed.
A Malayalam speech stack, English subtitles and a second visual style, the neon build reel.
Chapter labels, spec cards, live FPS overlays, thumbnail and post kit.
A 71-minute 4K two-camera session cut into two episodes, plus nine client-branded reels from a separate studio shoot.
V1Why it is hard to copy
Long 4K, multi-camera sessions run on our own GPU workstation with no upload caps, and client footage stays private.
On our benchmark, the best model read about 80% of Malayalam chunks correctly where stock Whisper failed. A proof sheet covers the rest.
Pacing, framing, card timing and loudness come from rules refined across twenty finished pieces.
Benchmark: one two-minute Malayalam and English test recording, split into chunks of seven seconds or less.
T1Roadmap
Only the first column ships today. The rest is plan.
A1For investors and partners
We finish your clients' content under your brand.
Our engine behind your podcast or editing product.
Send one real episode. We return a finished cut to judge.
Fund the step from proven workflow to product. Terms to be agreed.
Where we stand: the output is proven. The per-project scripts still need to become one reusable product, and parts of face tracking and rendering still run on CPU. That is what a partner unlocks.
T1Questions
Automatic video editing means software makes the editing decisions that a person would normally make by hand: which take to keep, where to cut dead air, when to show B-roll and where captions land. EDITSON does this from audio, face and word-timing signals, and a person reviews the result before delivery.
Yes. Captions are built from the final cut audio and timed word by word, then proofed by a person. We have shipped captions and subtitles in English and Malayalam, including speech that mixes the two.
English and Malayalam are proven in production. A dedicated Indian-language speech model covers 22 constitutionally recognised languages plus English, Bhojpuri and Bhili. Accuracy differs by language, so we benchmark yours on a real episode before committing.
Yes. It syncs multiple cameras and microphones, cuts to the active speaker, removes microphone bleed, picks a cold-open hook and writes chapters and titles. A 71-minute 4K two-camera session was turned into two finished episodes.
Yes. It produces vertical 9:16 edits with word-by-word captions, graphics cards, B-roll and a mastered soundtrack, and 16:9 films for YouTube. Both are part of our delivered work.
The cutting, captioning, graphics and audio mix run automatically. A person checks names, terms and the final play-through before anything is delivered, because accuracy matters more than speed.
No. Client footage is processed on our own GPU workstation and is not uploaded to third-party editing services.
On our own NVIDIA CUDA GPU workstation with 24 GB of video memory. It runs GPU speech recognition, subject matting and hardware video encoding, which lets it handle long 4K multi-camera sessions without upload limits.
Cost and turnaround depend on the length, languages and formats. The simplest way to find out is a pilot: send one real episode and we return a finished cut so you can judge the work.
Yes. Agencies and studios can use EDITSON under their own brand, and software teams can integrate the engine behind a podcast or editing product. Investment conversations are also open.
V1Next cut
We return a finished, post-ready cut so you can judge the work.