FastPix Encoding
Per-title encoding that tunes itself
FastPix is a video API for on-demand and live streaming. Every file you send it gets profiled and given a bitrate ladder built for that one video, so you ship the same picture in around 35% fewer bytes. No presets to write, no resolutions to declare, on by default.
$25 in signup credits. No credit card. Encoding is free on Standard quality.
# Upload a file. That is the whole configuration.
curl -X POST https://api.fastpix.io/v1/on-demand \
-H "Authorization: Bearer $FASTPIX_TOKEN" \
-d '{ "inputs": [{ "url": "https://cdn.acme.io/ep41.mov" }] }'
→ 201 Created
{
"playback_id": "aBc123",
"encoding": { "mode": "per_title" },
"status": "processing"
}Encoding video in production for
Per-title encoding builds a bitrate ladder for each video instead of reusing one across your catalog.
FastPix profiles every source you upload, works out how many bits that particular video needs, and publishes only the quality levels that earn their place. You get the same picture in around 35% fewer bytes.
Also called content-adaptive encoding. The alternative is a fixed ladder: one set of resolutions and bitrates applied to everything you own.
Smaller files, same picture
FastPix reads each source before encoding it, then spends bits where that particular video needs them. A flat interview and a fast-cut trailer come out of the same API with completely different ladders.
The target is a quality score, not a bitrate. The encoder aims at how the video will look and spends whatever it takes to get there, which is why the saving shows up without anything looking worse.
- Around 35% fewer bytes at matched quality, against open-source H.264, HEVC and AV1 baselines.
- Quality scored with VMAF before the asset is marked ready, not assumed.
- Every rung, not just the top one. The whole ladder is derived, including where the resolution changes.
A fixed preset would have sent 5.1 Mb/s at 1080p for this file. Low motion, flat background, so the ladder allocates less and the picture is identical.
Head to head
Per-title vs a fixed ladder: what actually changes
A fixed ladder is written once and applied to everything. We derive the rungs from each file instead. Here is what that changes, dimension by dimension, and then in measured numbers.
| Fixed ladder | Per-title | |
|---|---|---|
| How the rungs are chosen | Written once by hand, then reused | Derived from the source, per file |
| Content awareness | None. A cartoon and a football match get the same rungs. | Motion, detail and scene complexity decide the rungs |
| What it aims at | A bitrate you picked | A quality score, measured with VMAF |
| Simple content | Over-served, and held at low resolution longer than it needs to be | Higher resolution at the same bitrate |
| Complex content | Starved at the top rung | Bitrate raised where the picture needs it |
| A new content type | Needs a new preset, and someone to own it | Nothing to add |
| Encode cost | Lower per file | Higher per file. On FastPix it does not reach your bill. |
| Where the saving lands | — | Storage and delivery, every month for the life of the asset |
Measured, on two real titles
Netflix published both sides when they introduced the technique in 2015. The win arrives in two different shapes depending on the content, which is why one headline percentage never fits a whole catalog.
| Content | Fixed ladder | Per-title | What changed |
|---|---|---|---|
| Simple animation flat regions, low motion | 480p at 1,750 kbps | 1080p at 1,540 kbps | Four times the pixels, slightly less bandwidth. The win arrives as resolution. |
| Live-action drama top rung | 1080p at 5,800 kbps | 1080p at 4,640 kbps | Same resolution, 20% less bandwidth. Here the win arrives as cost. |
Figures published by Netflix in Per-Title Encode Optimization, December 2015. On FastPix the same technique runs on every asset with nothing to configure.
Nothing to tune, ever
There is nothing for you to configure: no resolution list, no bitrate table, no preset file to keep in sync with your content types. FastPix derives the ladder from the upload itself, so it is a property of the file rather than a decision you make in advance.
That matters most when your content changes. A new format, a new vertical, a screencast that behaves nothing like your demo footage: none of it needs a new profile, because the encoder was never following one.
- One request. A source URL is the entire job definition.
- No per-content-type profiles. The source decides, so new formats do not need new presets.
- Pin it if you must. You can fix codecs or a quality tier when a contract requires it.
The old way
On FastPix
The ladder keeps learning after you ship
Most encoders read the source file and stop there. FastPix keeps watching after playback starts, because Video Data is part of the same platform rather than a second product you buy.
Viewer Experience monitoring tracks errors, rebuffering, startup time and picture quality across millions of streams a day. Machine learning feeds those signals back into how ladders get built, so the encoder that runs your next upload is not the one that ran your last.
- Real sessions, real devices. Signals come from production playback, not lab conditions.
- Both failure modes. Too fat and viewers stall on weak connections. Too thin and the picture goes soft.
- Nothing to wire up. No integration, no configuration, no separate contract.
Signals in
Model
Ladder out
↻ your next upload starts here
How it works
Playback starts before the encode finishes
FastPix hands back a playback URL before the full ladder is done, so your first viewer is watching in seconds. Per-title runs inside Video on Demand and Live Streaming, same encoder, one API key.
Send the source
A direct upload, a pull from a URL, or a resumable upload from mobile. H.264, HEVC, AV1 or VP9 in, up to 4K with HDR.
FastPix profiles it
Motion, detail and scene complexity decide how many bits the picture actually needs, and where the resolution should change.
The ladder is built
Renditions in AV1, HEVC and H.264, packaged in CMAF with HLS and DASH manifests, scored against the source.
Playback starts
A playback URL comes back before the full encode finishes, so the first viewer is watching in seconds.
Background
What is per-title encoding?
A talking-head lecture and a football match do not need the same amount of data to look good. Most video libraries send them both out at the same settings anyway. Per-title encoding builds a bitrate ladder for each video instead of reusing one across your whole catalog. A ladder is the set of quality levels you publish, and each level in it is a rung; the player climbs up and down them as the viewer's connection changes. With a fixed ladder, someone on your team picked a set of resolutions and bitrates once and now owns them, and that one set has to survive your hardest footage. Everything simpler gets more bits than it can use.
On FastPix the ladder comes from the file, not from a profile you maintain. It runs on every asset by default. Netflix introduced the technique in 2015, and the chart shows why it works.
Each resolution has its own quality curve and the curves cross. Below a crossover, a clean 480p beats a blocky 1080p at the same bitrate. Those crossings sit in a different place for every video, which is the thing one fixed ladder cannot express.
Measured, not asserted
We publish the runs we lose too
Per-title encoding does more work per file than a fixed preset, because we profile the source before picking a ladder. FastPix still finishes first on most of the runs we publish, and the benchmark reports the files where it does not. Same file, same region, same network, method published alongside the numbers.
Runs a premium OTT catalog on FastPix across H.264, HEVC and AV1. Their delivered bytes are the reference workload behind the compression figure above.
Delivered bytes at matched visual quality, same catalog, same viewers. The saving lands on storage and delivery every month, not once.
What you get
On every asset, at no extra cost
Everything here ships on every FastPix asset, and we charge for none of it separately. Resolution and quality tier set your rate; rung count, codec and the analysis pass do not enter the calculation.
Per-title included. Your bill comes from storage and streaming only.
Higher rendition quality where picture is the product. Rates scale by resolution, not by rung count.
A twelve-rung AV1 ladder bills the same as a four-rung H.264 one. Only resolution and tier change the rate.
Common questions
Per-title encoding, answered
What is per-title encoding?
Per-title encoding builds a bitrate ladder for each video instead of reusing one across the catalog.
Netflix introduced the technique in December 2015. It solves the problem that a fixed ladder has to be provisioned for your hardest content, which means every simple video gets over-served and held at low resolution longer than it needs to be. Content-adaptive encoding, explained.
How does per-title encoding work?
The encoder analyses the source, then picks a resolution and bitrate for each rung.
It encodes the video across a grid of resolutions and quality settings, scores every result, and keeps only the encodes that nothing cheaper beats on quality. The ladder is a handful of points sampled along what survives. FastPix predicts where that sits from the content rather than discovering it by encoding the file hundreds of times.
How does per-title compare to a fixed ladder?
A fixed ladder applies one set of rungs to your whole catalog. Per-title derives them per video.
Because a fixed ladder has to survive your hardest content, it over-serves everything simple. On easy footage per-title tends to give you a higher resolution at the same bitrate, and on typical footage it gives you the same resolution for roughly 20% fewer bits.
Do I need to turn it on or configure anything?
No. Per-title runs on every FastPix asset by default.
There is no resolution list to declare, no bitrate table and no preset file. If a contract requires specific codecs or a quality tier you can pin those, but the ladder itself is always derived from the source. Five-minute quick-start.
How much bitrate does it actually save?
Around 35% fewer bytes at matched quality against open-source baselines.
The range is wide because it depends on your content. Simple footage saves far more than sport does, and on very easy content the gain often arrives as higher resolution at the same bitrate rather than a smaller file. The benchmark.
Keep reading
Where this fits
Per-title is one part of the FastPix encoding stack. The pages below cover the rest of what we run.