Does FastPix generate video?

No, and it will not. FastPix is the media layer either side of the model: store, deliver, play, automate, measure. You choose your own model and keep your own weights.

How do I ingest a clip without re-uploading it?

POST the source URL to /v1/on-demand. FastPix pulls from S3, GCS, Azure Blob or any public HTTPS link, encodes, and returns a playback ID. The file never passes through your servers.

Can I add captions to a generated clip automatically?

Yes. Set subtitles on the create call and FastPix generates them in the language spoken in the clip, at $0.048 per minute. A webhook fires when they are attached. FastPix does not translate them.

Can I normalize loudness on a video I already stored?

Not in place. Loudness is set on the create call and cannot be changed afterwards. To level something already stored, re-pull it with fp_mediaId://{id} and optimizeAudio on, which creates a new media entry with a new ID. Plan for that ID change in your database.

Do you dub or translate the audio on a generated clip?

No. FastPix attaches, replaces, overlays and levels audio you supply, and generates captions in the language spoken in the clip. It does not translate captions and it does not synthesise a dubbed track. Your voice model produces the language, FastPix carries it.

Does moderation stop a video from going live?

No, and no video platform should let it silently. Create the asset private, read the moderation event, then flip access once your own rule clears it. Scoring is sampled across 13 categories rather than frame by frame, and there is no threshold setting, so the decision stays in your code.

Do you support C2PA or content credentials?

No. There is no provenance signing and no invisible watermarking. Visible watermarking is available, which is a different thing and does not satisfy a provenance requirement. If you have an EU AI Act transparency obligation, that layer sits elsewhere in your stack.

What does a five second take actually cost?

Five seconds of encode, plus whatever is delivered and stored. There is no per-title fee, so a generation nobody opens costs close to nothing, and deleting it stops the storage line. See pricing.

Can I prune old generations automatically?

Not from FastPix today. There is no TTL or lifecycle policy, so pruning means your own scheduled job calling delete on assets past a threshold. The delete call itself is one line and immediate.

What happens when a product goes viral overnight?

Delivery is multi-CDN and billed per minute, so a spike costs more but does not need a plan change or a call. Encoding queues on short clips are the part worth load-testing before a launch.

Can I tell which model version people actually watch?

Tag each asset with model version and prompt style using custom metadata, then read completion and replays against those fields. FastPix returns your metadata on every read but does not index it, so the joining happens in your own analytics.

Can I watermark the free tier and not the paid one?

Yes. A visible image or text watermark is a transform applied to the assets you choose, placed by pixel or percentage with opacity control. Pair it with an MP4 rendition for the download button and a signed link for paid output.

FastPix for generative AI

Video API for generative AI apps

One API to store, encode, deliver and publish the video around any generative model, so your team spends its time on the model instead of on video infrastructure.

Start building Get a demo

Per-minute pricing, no per-title fee and no per-seat fee. $25 in signup credits, no credit card.

Storeassets in, output backPlaypreview to exportMeasureQoE and replays

Trusted by teams shipping video at scale

Why FastPix

Complete video infrastructure for generative AI products

One API, whole stack

Storage, encoding, playback, watermarking and data all come from the same API key.

Built for scale

Multi-CDN delivery scales with demand, so a launch spike runs on the same setup as everyday load.

Fast by default

Output lands and starts playing quickly, which matters when a user has already watched a queue counter.

Pay per minute

Billed on minutes, so short generated clips cost like short clips. Docs, SDKs and engineers in Slack.

Developer reference

Every endpoint a generative AI product uses

Everything on this page maps to a documented call, grouped the way a generation moves through your pipeline.

Ingest and manage

Upload media from device Create media from URL Upload from a URL Get media input info Add metadata to videos Delete a media by ID Webhook: media events Video on demand guide

Encode and transform

Configure quality levels Clip and trim videos Add a watermark to a video Add an intro and outro Merge and stitch videos Create GIFs from a video Webhook: transform events Thumbnails and hovers guide

Audio and moderation

Add audio to a video Replace a video's audio track Overlay audio on a timeline Normalize audio loudness Detect NSFW and profanity Extract named entities Webhook: in-video AI events In-video AI guide

Access and export

Generate JWTs for secure media Secure media access with JWTs Restrict playback access Update domain restrictions Create a playback ID Update source access Webhook: set up and verify Signed playback guide

Players and SDKs

Player SDKs overview Install the web player iOS player Android player Preload and CORS settings Handle playback errors Webhook: player events Player SDK guide

Analytics

Video Data overview Quality of experience metrics Video startup metrics Use custom dimensions Troubleshoot playback errors Build workflows with the API Webhook: exception alerts Video Data guide

Assets

Store and deliver AI-generated video at scale

Users upload reference frames. Your model reads them and writes clips back. FastPix holds both ends, each with a URL your model can fetch and your own fields attached.

Resumable upload from the browser for reference frames and clips, and URL pull for whatever your model writes. Guide
Ten custom fields per asset: user, project, model version, prompt hash, favourite. Guide
Delete when a user clears history and the storage line stops with it. Guide

Transformations

Video transformation API for clips, GIFs and watermarks

Your model writes one file, but the product still needs thumbnails, GIFs and watermarks. FastPix runs those as transforms on the stored asset, so your team stays on output quality.

Clip and trim to a share length, with no re-upload and no second encode. Guide
Stills, GIFs and timeline hovers, so a gallery previews on hover instead of sitting static. Guide
Watermark by pixel or percentage with opacity control, on the assets you choose. Guide
Intro and outro stitched on, so an export carries your brand without a render step. Guide

Merge and stitch guide

Audio

Audio API to add, replace and normalize a soundtrack

A generation comes back silent, or with audio that is louder than the last one. Add the track, place it where you want it, and set every clip to the same volume.

Add an audio track to a silent generation, the narration or score your voice model returned. Guide
Replace the audio track for a re-recorded take or another language, without touching the video. Guide
Overlay audio on the timeline, placing a cue at a point in the clip rather than across all of it. Guide
Audio normalization, so a quiet take and a loud one both end up at the same volume. Guide

Audio and subtitle tracks guide

Moderation

Content moderation for AI-generated video

A model will occasionally return something you cannot publish, and an open prompt box is where it starts. FastPix samples frames and audio, scores the result and posts it to a webhook.

NSFW, hate speech and profanity detection across 13 categories, read from sampled frames and the audio track. Guide
Named entities pulled from speech, so a brand or person the output mentions reaches your own index. Guide
One event carries the verdict, so the review queue and the publish step read the same signal. Guide

In-video AI guide

Playback

Video player SDK for previews, shares and exports

Your user already watched a queue counter, and a second spinner is where it breaks. The player mounts on a playback ID and starts on the first rendition, share link and download included.

React component, iframe or native views, with scrubber, mute and fullscreen already in it. Guide
Watermark the free tier, clean file when they upgrade. One transform, not a renderer change. Guide
MP4 renditions for the download button, and signed links when the output is paid. Guide

Pipeline automation

Automate the video pipeline around your model

Everything after the model is a flag or a webhook. Set captions, a poster frame and a watermark on the create call. An event fires as each finishes, so nothing polls.

Captions on output headed for social, in the language spoken in the clip. Guide
Poster frame, hover loop and watermark set at create time, not as jobs you queue afterwards.
A webhook per step, so your UI flips from Creating to playing without a poll loop. Guide
Transform events tell you when a watermark or a trim finished on an asset you already published. Guide

Video data and QoE

Video analytics and QoE monitoring for every play

Two questions you cannot answer without playback data. Is the video reaching people cleanly. Is the output any good. Startup and rebuffering answer the first, replays against your model version answer the second.

QoE per session: startup time, rebuffer ratio and errors by device and region, across 56 dimensions.
Replays per take, read against the model version you tagged at ingest. A better signal than a rating widget.
100,000 session plays a month included, then $299 a month for a million.

Explore Video Data

API and SDKs

Video API and SDKs for a generative AI stack

Generative products are built by small teams who did not plan to become video engineers. Every step above is one call, and the SDK is in the language your pipeline already speaks.

SERVER SDKS

Node Python Java Go Ruby PHP .NET

CLIENT SDKS

iOS Android Flutter React Native Web Player

Read the API reference

the whole integration

POST/v1/on-demandPull the generation

GET/thumbnailGallery still

GET/gifHover loop

POST/watermarkBrand the output

GET/signing-keysGate a paid tier

GETVideo Data APIPer-model performance

DELETE/on-demand/:mediaIdPrune cold assets

HOOKmedia eventsDrive the pipeline

Benchmarked against five platforms

Upload and startup, measured against five platforms

Generated clips are short, plentiful, and a user is already waiting. A heavy 4K file and a short clip were pushed through six platforms and measured on ingest and startup time.

Upload time245 MB 4K package, lower is better

FastPix20.5s

Platform A31.9s

Platform B111.7s

3.5x faster than the average across platforms tested. Two others could not take the 4K file at all: one caps uploads at 100 MB, the other errored in processing.

Startup timeshort clip, cold start to first frame

FastPix1.0s

Platform C1.7s

Platform A2.8s

1.0s, the fastest of the six and well inside the 2 second abandonment threshold. A seventh reading came back as 0ms and was excluded as invalid.

Multiples are measured against the average across platforms tested. Per-title encoding tunes the ladder to the source, which is why a phone held sideways in poor light still gets a usable rendition.

How it comes together

Build the video pipeline behind a model like Veo or Kling

Five steps between a prompt and a clip a user can watch and share.

Store the reference

Take the image or clip a user supplies and hand your model a URL to fetch it from.

Guide

Take the output back

Pull the generated file in from a URL and encode it once for every screen.

Guide

Fix the sound and the frame

Attach the voice track, level the loudness and set the poster frame so every take matches.

Guide

Play it straight away

The player mounts on a playback ID and starts on the first rendition, no second spinner.

Guide

Watch the queue

Events fire as each step finishes, so your job queue moves without polling for status.

Guide

Explore AI video workflows Explore video APIs in the AI era

Tell us where video sits in your model pipeline

Walk us through your architecture and where video has to plug in. We will map it to the API surface and tell you which parts are one call and which need real work.

· 30 minutes· Bring your pipeline· Engineers, not slides

Talk to us

Prefer to build it yourself? Start building free

Common questions

Generative AI video questions, answered.

Does FastPix generate video?
No, and it will not. FastPix is the media layer either side of the model: store, deliver, play, automate, measure. You choose your own model and keep your own weights.
How do I ingest a clip without re-uploading it?
POST the source URL to /v1/on-demand. FastPix pulls from S3, GCS, Azure Blob or any public HTTPS link, encodes, and returns a playback ID. The file never passes through your servers.
Can I add captions to a generated clip automatically?
Yes. Set subtitles on the create call and FastPix generates them in the language spoken in the clip, at $0.048 per minute. A webhook fires when they are attached. FastPix does not translate them.
Can I normalize loudness on a video I already stored?
Not in place. Loudness is set on the create call and cannot be changed afterwards. To level something already stored, re-pull it with fp_mediaId://{id} and optimizeAudio on, which creates a new media entry with a new ID. Plan for that ID change in your database.
Do you dub or translate the audio on a generated clip?
No. FastPix attaches, replaces, overlays and levels audio you supply, and generates captions in the language spoken in the clip. It does not translate captions and it does not synthesise a dubbed track. Your voice model produces the language, FastPix carries it.
Does moderation stop a video from going live?
No, and no video platform should let it silently. Create the asset private, read the moderation event, then flip access once your own rule clears it. Scoring is sampled across 13 categories rather than frame by frame, and there is no threshold setting, so the decision stays in your code.
Do you support C2PA or content credentials?
No. There is no provenance signing and no invisible watermarking. Visible watermarking is available, which is a different thing and does not satisfy a provenance requirement. If you have an EU AI Act transparency obligation, that layer sits elsewhere in your stack.
What does a five second take actually cost?
Five seconds of encode, plus whatever is delivered and stored. There is no per-title fee, so a generation nobody opens costs close to nothing, and deleting it stops the storage line. See pricing.
Can I prune old generations automatically?
Not from FastPix today. There is no TTL or lifecycle policy, so pruning means your own scheduled job calling delete on assets past a threshold. The delete call itself is one line and immediate.
What happens when a product goes viral overnight?
Delivery is multi-CDN and billed per minute, so a spike costs more but does not need a plan change or a call. Encoding queues on short clips are the part worth load-testing before a launch.
Can I tell which model version people actually watch?
Tag each asset with model version and prompt style using custom metadata, then read completion and replays against those fields. FastPix returns your metadata on every read but does not index it, so the joining happens in your own analytics.
Can I watermark the free tier and not the paid one?
Yes. A visible image or text watermark is a transform applied to the assets you choose, placed by pixel or percentage with opacity control. Pair it with an MP4 rendition for the download button and a signed link for paid output.

Pricing

Rate card

Per minute, with the player and analytics already in it. No per-title fee and no per-seat fee. See full pricing.

ENCODING

Pay per minute encoded

$0.0267/min 1080p Pro

Per-title ABR encoding across H.264.

Up to 720p Pro$0.0213 / min

Up to 1080p Pro$0.0267 / min

1440p (2k) Pro$0.0427 / min

2160p (4k) Pro$0.0855 / min

STREAMING

Pay per minute streamed

$0.00067/min 1080p Pro

Multi-CDN delivery from 300+ POPs.

Up to 720p Pro$0.00054 / min

Up to 1080p Pro$0.00067 / min

1440p (2k) Pro$0.00108 / min

2160p (4k) Pro$0.00217 / min

STORAGE

Pay per minute stored

$0.00268/min 1080p Pro

Assets stored and CDN caching, lifecycle handled.

Up to 720p Pro$0.00215 / min

Up to 1080p Pro$0.00268 / min

1440p (2k) Pro$0.00430 / min

2160p (4k) Pro$0.00860 / min

See full pricing Why FastPix →

Support

Three ways to get unstuck

Engineering

Engineering support

Stuck on an API call, a webhook signature or a player integration? Reach the engineering team directly, and hear back in hours rather than days.

Contact engineering

Docs

Integration help

Quickstart guides, SDK examples and playback logs in your dashboard, for the integrations most teams hit first.

Browse the docs

Community

Solution architect

A new build, a migration off another platform or a complex multi-tenant rollout. Plan it with a FastPix solution architect.

Join the Slack community