Find the over-delivery
Score your current encodes against the source. See which titles sit well above your quality bar, and what those extra bytes cost.
FOR STREAMING OPERATORS & BROADCASTERS
Cut video delivery costs at the quality you choose.
When every title is encoded with the same settings, many are delivered with more data than your quality standard requires. hls.ai finds the smallest encode that still meets your standard, title by title, and verifies it inside your infrastructure.
VOD & catch-up · Runs in your environment · Start with a few titles
ONE QUALITY TARGET. FEWER BYTES.4 TEST CLIPS
Baseline · H.264, fixed settingQuality target · HEVC, mean VMAF 95
Measured on complete clips.VIDEO DATA
58%
less video data across four test clips
HEVC at a VMAF 95 quality target vs. fixed-setting H.264
Per clip: 44–61% against H.264, 40–59% against fixed-setting HEVC.
Your result depends on your content. Test it on your own footage.
WHERE DELIVERY COST HIDES
Are you delivering more data
than your quality bar needs?
Different titles need very different amounts of data to look the same. A single encoding recipe leaves easy titles far above your quality standard, and you pay to deliver the difference on every play. hls.ai measures that gap on your content and removes it.
Encode to your quality bar
For each title, find the smallest encode that still meets the target you set, with a floor for the hardest scenes. The output is standard video; no special player.
Verify before you commit
Every result comes with side-by-side video, frame-level quality scores, and processing time, so your team can check the trade-off.
01 / THE APPROACH
Your quality bar.
The smallest encode that meets it.
You set the standard. hls.ai searches for the right setting for each title, then checks the result against the source.
Set the target
Choose the quality score your service must hold, and a floor for the hardest scenes. Your current encodes are the baseline.
Search each title
A long title is searched on a few short samples, by default four of 5 seconds, not re-encoded in full. Trial encodes cover only those samples, and the full title is encoded once at the end. Our test clips needed five or six trials each.
Verify the output
The finished encode is scored against the source, frame by frame. You get the bytes, the scores, and the frames that came closest to the floor.
LONG-FORM
Long titles: optimized chunk by chunk, not once per title.
A two-hour title has easy stretches and hard ones. With one setting for the whole title, the hardest few minutes decide the bitrate for all of it. So for long-form content we optimize chunk by chunk: each stretch gets the smallest encode that meets your quality bar. Keyframes stay on the schedule your HLS packager already uses, so your segments and players stay as they are.
The clips on this page are short single shots, so the results shown are per-title. Chunk-level results are measured on your own long-form titles during an evaluation.
The published results use HEVC at the top rendition and native resolution. Codec, settings, and the rest of your ladder are agreed for each evaluation, with every rendition on the same keyframe schedule. Compute cost per title-hour is measured on your content. Our preprocessing research, on stable-background reuse and per-area frame rates, is in the technical evaluation.
02 / REAL FOOTAGE, MEASURED RESULTS
Set the quality target.
Send fewer bytes.
Encoding to a quality target instead of a fixed setting cut video data by 44–61% on four clips. Watch the comparison first.
BASELINE · H.264QUALITY TARGET · HEVC
First 8 of 18.5 seconds · Resized side-by-side preview · SilentFocus on the face, hands, and moving edges.
Baseline video data
10.096 MBQuality-target video data
4.218 MBMeasured data reduction
58.2%Video data avoided per play
5.877 MBStudio talk, complete 18.5-second clip at 1080p. Baseline: H.264 medium CRF 20. Quality target: HEVC, with the setting selected automatically to keep mean VMAF at 95 or higher. Mean VMAF: 95.88 → 95.56.
Four clips, one quality target
The target for every clip: mean VMAF of 95 or higher, with 95% of frames at 92 or higher. The baseline is H.264 medium CRF 20. The fixed-HEVC column compares with HEVC at one setting for every clip (slow, CRF 20).
| Clip | Length · resolution | Baseline · H.264 | Quality target · HEVC | vs. H.264 | vs. fixed HEVC | Mean VMAF | Compare |
|---|---|---|---|---|---|---|---|
| Studio talk | 18.5 s · 1920 × 1080 | 10.096 MB | 4.218 MB | 58.2% saved | 47.6% saved | 95.88 → 95.56 | Watch |
| Basketball | 26.4 s · 2048 × 1080 | 25.089 MB | 10.457 MB | 58.3% saved | 53.3% saved | 97.44 → 95.15 | Watch |
| Sword action | 5.6 s · 1920 × 1080 | 9.060 MB | 5.112 MB | 43.6% saved | 39.9% saved | 97.36 → 96.09 | Watch |
| Fireworks | 13.8 s · 2048 × 1080 | 34.154 MB | 13.198 MB | 61.4% saved | 58.5% saved | 99.87 → 97.39 | Watch |
| All four clips | 64.3 s | 78.399 MB | 32.985 MB | 57.9% saved | 53.4% saved | — | Details |
This is not an equal-quality comparison: the quality-target encodes score 0.3 to 2.5 VMAF points below the baseline. The H.264 figure combines three effects: the switch from H.264 to HEVC, a slower encoder preset, and stopping at the quality target. The fixed-HEVC figure isolates the last one: same codec and preset, with only the setting chosen per title. Its baseline scores VMAF 97 to 99.9, so it is not an equal-quality comparison either. These are standard encoder controls; hls.ai automates choosing and verifying them. VMAF is a quality metric developed by Netflix, and it is an automated estimate, so judge the footage yourself.
Codec comparisons and our preprocessing research, including the experiments that did not help, are in the technical evaluation.
03 / ENTERTAINMENT CATALOGS
Different scenes.
Different treatments.
Start with representative titles from your catalog. Test each content type against your existing encodes, and select treatments only where quality and savings hold up.
Drama & film
Evaluate dialogue, dark scenes, camera movement, and action separately. Preserve faces, texture, and intended film grain while testing where fewer bytes are possible.
Anime
Flat colour and held frames may need far less data than live action. Check line art, subtitles, and intentional animation timing at each quality target.
Variety & sports
Compare studio segments, on-screen graphics, and fast action. Protect text and motion; keep the existing treatment when an alternative does not help. Start with recorded and catch-up content.
ROUGH ESTIMATE
What would fewer bytes
be worth to you?
Enter your own delivery figures. This is arithmetic on your numbers, not a forecast. An evaluation measures the real reduction on your content.
Our test clips showed 44–61% against a fixed-setting H.264 baseline; a pipeline that is already tuned will see less. Assumes delivery billed by volume. Processing cost is not subtracted.
04 / ENTERPRISE PRICING
Your infrastructure.
Our optimization software.
Our primary offer is software running in your environment. Managed processing is a higher-priced option for teams that want us to operate the workflow. Deployment scope and operational requirements are proposed after evaluation.
Evaluation deliverables include MP4 comparison videos and measurements. Production integration, managed workflows, and HLS packaging are scoped and validated separately before deployment.
01 / EVALUATION
Benchmark your content
Scope agreed together
Compare representative footage with your existing encoding pipeline before committing to a production project.
- Representative sample selection
- Side-by-side quality assessment
- Measured bytes and estimated cost impact
02 / PAID PILOT
Validate at catalog scale
Fixed project quote
A defined processing project with agreed content, quality targets, deliverables, and evaluation criteria.
- Multiple content types and resolutions
- Integration and processing cost assessment
- Verified pilot results before expansion
03 / PRODUCTION
Optimize your workflow
Custom annual agreement
Receive a tailored proposal covering the software license, processing volume, and support.
- Volume-based processing terms
- Deployment requirements agreed in advance
- Optional fee linked to verified net savings
What determines the quote?
Content minutes, resolution, encoding outputs, processing intensity, deployment requirements, and support. We assess savings using your actual delivery rates and subtract additional processing and storage costs before discussing a performance fee.
The results above are reductions in video bytes on test clips, not a guaranteed reduction in your streaming bill. Production scope depends on validated quality, economics, and deployment requirements.
Preparing for an evaluation
Useful starting information: representative content you’re permitted to share, current codec and encoding settings, monthly delivered video volume, your billing model, and quality requirements. These let us assess whether reducing bytes could lower your costs.
Services billed by viewing minutes may not directly save money from smaller files. A benchmark should establish both technical improvement and a financial benefit before a production agreement.
WHY WE BUILD THIS
Built by engineers
who paid the bandwidth bill.
TEAMSho Shimoda (CEO, development), Yasushi Motoki (product development, implementation support), Koichi Nin (development, support)
TOOLSFFmpeg, x265, and libvmaf. Standard tools combined, not a proprietary codec.
COMPANYRECEIPTROLLER Inc., founded in 2022 in Nagano, Japan
Company site
TRY IT IN YOUR ENVIRONMENT
See what your
catalog could save.
Run the evaluation inside your own environment, so your content never has to leave it. If sharing is easier, a few representative titles are enough. Either way, your team gets comparison videos and measurements it can check.