Fivelabs: AI video studio for narrated Shorts and long-form videos

Fivelabs: Fivelabs is an early-stage company building AI tools that turn writing into finished video.
Fivelabs Studio: Fivelabs Studio writes or takes your script, films every beat, narrates it and cuts it into a Short or a long-form video you can edit and publish.
Our render stack: We run the open-weights MiniMax H3 video model on GPUs rented from a decentralised network, so a video costs what its GPU minutes cost.

Shorts and long-form, from one studio.

Same writer, same voices, same editor. Pick the format and the studio plans the shots, the images and the motion to fit it.

Vertical Shorts, 30–60 seconds

“Why Ganesha broke his tusk”: Hindi narration, English captions, 28 seconds.

Every shot is a video clip from MiniMax H3, with the model’s native audio under the narration.
Upright by design. Keyframe-first portrait rendering keeps vertical video the right way up, and every take passes an automatic orientation review.
Retakes and retries. Re-roll any shot you don’t like; failed renders retry automatically.
Captions burned in with your choice of style, highlight and position.

A colonial-era mansion among palms with the chapter title ‘A house called joy’. — Frames from a five-minute long-form video on the history of Anand Bhawan, made in Fivelabs Studio.

A grand dining room with a chandelier and high windows. — Frames from a five-minute long-form video on the history of Anand Bhawan, made in Fivelabs Studio.

Long-form videos, 1–20 minutes

Chapters with title cards, planned from your script.
AI images with smooth Ken Burns motion, plus moving images animated from the stills.
19 motion-graphics templates for numbers, timelines, maps, quotes and charts.
Cinematic intros and avatars, with lip-sync from a specialist provider.
Music, captions and film-look grading on the final cut.

More Shorts from the studio

The secret of 18 in the Mahabharat
Mahabharat: the longest epic
One grain of rice that stopped a curse
The history of Ghaziabad

Four steps from idea to export.

The studio shows a live credit estimate at every step, so you know what a video will cost before anything renders.

Idea or script

Type a topic or paste your own script, and describe the visual direction you want.
Narrator and length

Choose a language, one of 33 narrators or a natural or custom voice, and how long the video should run.
Style and upgrades

Pick one of 15 image styles and add upgrades such as motion graphics, a cinematic intro or an avatar.
Review and generate

Approve the script, generate, polish it in the editor and export in 1080p.

Prefer code or chat? Every step is also available through the REST API and the MCP server.

Stories that answer their own hook.

The writer, built on Claude, drafts hooks and a script to a fixed shape. A check rejects drafts whose payoff never arrives or whose ending leaves the hook open.

Hook
Context
Turn
Reveal
Payoff
Close

Each draft comes with writer’s notes: the question the hook raises, how the payoff answers it, what the last line has to do, and the sources behind the facts. Nothing renders until you sign off.

The lake that killed four Pandavas

A script from the studio’s writer, for a Hindi Short with English captions.

Hook
Water from one lake, and four Pandavas fell lifeless. Why?
एक झील का पानी, और चार पांडव बेजान होकर गिर पड़े, आख़िर क्यों?
Context
In exile, the thirsty Pandavas found a clear lake while searching for water.
Turn
An unseen Yaksha said: answer my questions first, then drink.
Turn
Nakula, Sahadeva, Arjuna and Bhima ignored him, drank, and fell.
Reveal
Finally Yudhishthira came, and patiently answered every question.
Reveal
The greatest wonder? People die every day, yet everyone thinks they will live forever.
Payoff
Offered one brother back, Yudhishthira chose Nakula, so that Madri too would have a living son.
Close
Pleased by this fairness, the Yaksha, Dharma himself, revived all four: where strength failed, patience won.
जहाँ ताक़त हारी, वहाँ धैर्य जीत गया।

Writer’s notes

Hook question: Why did four Pandavas fall lifeless at one lake?
Payoff: They ignored the Yaksha’s questions; Yudhishthira answered them and chose fairly, so Dharma revived all four.
Ending: Brings the four back to life, resolving the hook, and lands the takeaway: patience over strength.
Sources: Vana Parva, Aranya sub-parva (Yaksha Prashna)

Waiting for your sign-off. Shots render after you approve the script.

Everything it takes to finish the video.

The studio covers the whole job, from the first line of the script to the upload.

EditorFix one shot without starting over.

Timeline with trim, replace and reorder
Crossfades and your own music upload
Caption styles with highlight, all-caps and position
Undo, redo and version history
Edit a section’s script and re-record just that section
Four thumbnail concepts per video: pick one or edit it

Thumbnail concept 1: Ganesha statue with the Hindi title ‘Why did Ganesha break his tusk?’.

Thumbnail concept 2: the sage Vyasa with the Hindi title ‘A broken tusk became a pen’.

Thumbnail concept 3: the sage in lamp light with the Hindi title ‘Vyasa spoke, Ganesha wrote’.

Thumbnail concept 4: a tusk on manuscript pages with the Hindi title ‘Who wrote the Mahabharata?’.

Voices and languagesHindi first, and six more.

33 curated narrators with previews, search and a gender filter
Natural voices from ElevenLabs
Custom voices saved to your account
Hindi, English, Spanish, Portuguese, French, German and Italian
Hindi narration with English (Hinglish) captions, or none
Caption timing taken from the narration itself

Visual styles15 image styles, one consistent look per video.

Photoreal
Cinematic
Natural
Painterly
Cartoon
Anime
Comic art
Isometric
Watercolour
Line drawing
Graffiti art
Pixel art
Oil painting
Neon art
Cubism

RenderingOur own GPU render stack.

MiniMax H3, an open-weights video model with native audio
Keyframe-first portrait rendering for upright vertical video
Automatic orientation review of every portrait take
Retakes per shot, automatic retries with backoff
1080p master export, plus a web encode and thumbnail
Credits meter real GPU cost, with a monthly budget

PublishingFrom export to channel.

YouTube uploads, labelled as synthetic media
Instagram Reels
An email when your video is ready
Signed, expiring links to share a cut

Video at GPU cost, not per-second API prices.

We run the open-weights MiniMax H3 model on our own render stack, on GPUs rented by the hour from Nosana’s decentralised network. A video costs what its GPU minutes cost.

Cost of one second of generated video

MiniMax H3 shots, US dollars. Lower is better.

Hosted MiniMax H3 API, per-second price$0.08
Fivelabs render stack, measured≈ $0.0025

$0.015 for a 6-second clip, divided by 6.

≈ 32×lower GPU cost per second of video than the per-second API price.

Worked examples
Output	API price	Our GPU time
One 6-second clip	$0.48	≈ $0.015
60 seconds of video clips	$4.80	≈ $0.15

Measured on an RTX A6000 rented on Nosana at about $0.36 per hour. A 6-second, 480×864 clip takes about 2.5 GPU minutes: 2.5 ÷ 60 × $0.36 ≈ $0.015. Clips are upscaled to 1080p in assembly.
API price: about $0.08 per second of video for the hosted MiniMax H3 API. Resolution and settings differ between the two, so read this as an order-of-magnitude comparison.
GPU time only. Narration, writing, storage, idle GPU time and engineering are not included.

Open weights, our own stack

Because we run the model ourselves, price follows GPU time rather than a vendor’s per-second rate, and we can tune steps, resolution and retries per shot.
Decentralised compute

We rent GPUs by the hour on Nosana instead of owning hardware. We don’t run data centres; we run the software that keeps rented GPUs busy with useful frames.
Long-form stays cheap too

The visuals for a 5-minute long-form video cost about $0.20 in GPU time. Long-form mixes AI stills with Ken Burns motion, moving images and motion graphics, so it isn’t a like-for-like comparison with per-second video.
Credits you can read

One credit is $0.01 of GPU or provider spend. You see an estimate before you generate, queued work reserves its estimate, and renders stop at your monthly budget.

Make videos from code, or from a chat.

REST API v1
Token-authenticated endpoints to create, edit, generate and render Shorts and long-form videos, with request IDs and consistent error codes.
MCP server with 26 tools
Lets AI assistants such as Claude and ChatGPT create videos for you. Apps connect through OAuth 2.1 with PKCE and an explicit consent screen, or with an MCP key.
Publishing built in
Send finished videos to YouTube and Instagram from the same workflow.

Developer overview Request API access

Create a ShortREST API v1

# Send "script", or "prompt" to let the writer draft it.
curl -s "$FIVELABS/api/v1/faceless-shorts/create" \
  -H "Authorization: Bearer $TOKEN" \
  -H "content-type: application/json" \
  -d '{
    "name": "Octopus hearts",
    "duration": 60,
    "narrator": "andrew",
    "imageTheme": "cinematic",
    "script": "Octopuses have three hearts. ..."
  }'

# Then: generate, render, poll the status endpoints.

Ask from Claude or ChatGPTMCP

You: Make a 45-second Short in Hindi about why Ganesha
     broke his tusk, cinematic style.

Tools: create_faceless_shorts
       generate_faceless_shorts
       get_faceless_shorts_rendering_status

Built to be trusted with your channel.

The studio handles your scripts, your voice, your budget and your publishing accounts, so we had it independently audited.

Independent security audit
A separate review covered sign-in, uploads, spending limits, the API and MCP. Findings were fixed with regression tests; open items are tracked.
90+ automated tests
They run against a fake GPU and mocked providers, so testing never touches real renders or real spend.
Hashed, revocable keys
API and MCP keys are stored hashed, can expire, and can be revoked at any time.
OAuth 2.1 for AI apps
PKCE, a consent screen that names the app and where it redirects, single-use codes and refresh-token rotation.
Spending you control
An estimate before every charge, reserved budget for queued work, and a hard stop at your monthly limit.
Careful media handling
Signed, expiring media links; uploads checked by their content, not their file name; image links fetched only from public hosts.

For investors, accelerators and programs.

Fivelabs is an early-stage company. The studio is built and working, and we’re raising and applying to accelerator, grant and inception programs.

Investor and program enquiries Read the one-pager

Stage: Early stage, actively building.
Product: Fivelabs Studio: writer, render stack, editor, long-form tools, voices, API and MCP, all shipped.
Edge: Open-weights video on rented decentralised GPUs: about $0.015 of GPU time for a 6-second clip that costs $0.48 at per-second API pricing.
Built for: Creators, educators and media teams, starting with Hindi and six more languages.
Looking for: Investment, accelerator places, GPU and cloud credit programs, and distribution partners.

Questions, answered plainly.

Something missing? Write to hello@fivelabs.xyz.

What is Fivelabs?

Fivelabs is an early-stage company that builds Fivelabs Studio, an AI video studio. It turns an idea or a script into a finished, narrated video: vertical Shorts of 30–60 seconds or long-form videos of 1–20 minutes, exported in 1080p.

How does Fivelabs Studio work?

There are four steps. Start from an idea or a script and describe the visual direction; choose a narrator and length; pick one of 15 image styles and any upgrades; then review the script and generate.

An AI writer drafts the script with a fixed story structure, you approve it, the MiniMax H3 video model renders each shot on our GPU render stack, and you finish the video in the editor and export it in 1080p.

Which languages does Fivelabs Studio support?

Narration in Hindi, English, Spanish, Portuguese, French, German and Italian. Hindi videos can carry English (Hinglish) captions or no captions.

How much does a video cost to make?

Inside the studio, 1 credit equals $0.01 of GPU or provider spend, and you see an estimate before anything is charged.

On our measured setup a 6-second, 480×864 clip takes about 2.5 GPU minutes on an RTX A6000 at about $0.36 per hour, which is roughly $0.015, and the visuals for a 5-minute long-form video cost about $0.20 in GPU time. The hosted MiniMax H3 API is about $0.08 per second of video, or $0.48 for the same 6 seconds.

Customer pricing is not published yet; ask us for a demo.

Why is Fivelabs cheaper per second of video than video APIs?

We run the open-weights MiniMax H3 model on our own render stack and rent GPUs by the hour on Nosana, a decentralised GPU network, instead of paying a per-second API price. Cost follows the GPU minutes a video actually uses. We do not own data centres or physical GPUs.

Can I create videos from code or from ChatGPT or Claude?

Yes. Fivelabs Studio has a REST API (v1) with token authentication for creating, generating and rendering Shorts and long-form videos, and an MCP server with 26 tools and OAuth 2.1 so AI assistants such as Claude and ChatGPT can create videos. It can also publish to YouTube and Instagram. See the developer overview.

Can I edit a video after it is generated?

Yes. The editor has a timeline with trim, replace and reorder, crossfades, music upload, caption styles, undo and redo, per-section re-recording, and four thumbnail concepts per video that you can pick or edit.

Is Fivelabs Studio secure?

The studio has had an independent security audit, and fixes are covered by more than 90 automated tests. API and MCP keys are stored hashed and can expire or be revoked, AI apps connect through OAuth 2.1 with PKCE and a consent screen, media links are signed and expire, and spending stops at a hard monthly budget.

What stage is Fivelabs at?

Fivelabs is an early-stage company. The studio is built and in use, and we are raising and applying to accelerator, grant and inception programs. Investors and program teams can write to hello@fivelabs.xyz.

Let’s make the first one together.

Tell us what you want to make, build or back.

See a demo

For creators, educators and media teams. We’ll walk through the studio with a topic of yours.
Request a demo
Partner with us

For publishers, platforms, compute providers and anyone building on the API or MCP server.
Partner with us
Invest or support

For investors, accelerators, and grant, credit and inception programs.
Investor and program enquiries

Or write directly to hello@fivelabs.xyz.

Turn a script into a finished, narrated video.

Fivelabs in three sentences

Shorts and long-form, from one studio.

Vertical Shorts, 30–60 seconds

Long-form videos, 1–20 minutes

More Shorts from the studio

Four steps from idea to export.

Idea or script

Narrator and length

Style and upgrades

Review and generate

Stories that answer their own hook.

Everything it takes to finish the video.

EditorFix one shot without starting over.

Voices and languagesHindi first, and six more.

Visual styles15 image styles, one consistent look per video.

RenderingOur own GPU render stack.

PublishingFrom export to channel.

Video at GPU cost, not per-second API prices.

Open weights, our own stack

Decentralised compute

Long-form stays cheap too

Credits you can read

Make videos from code, or from a chat.

REST API v1

MCP server with 26 tools

Publishing built in

Built to be trusted with your channel.

Independent security audit

90+ automated tests

Hashed, revocable keys

OAuth 2.1 for AI apps

Spending you control

Careful media handling

For investors, accelerators and programs.

Questions, answered plainly.

Let’s make the first one together.

See a demo

Partner with us

Invest or support