Skip to content
TubePress — free, self-hosted & actively maintained
All guides

Video Encoding for Self-Hosted Tube Sites: FFmpeg, AV1 & Storage Math

Guide 11 min read Updated Jul 2026
Video Encoding for Self-Hosted Tube Sites: FFmpeg, AV1 & Storage Math

Key takeaways

  • Three renditions cover a tube site (240/480/720); 1080p is a monetization lever, not a default.
  • CRF 23 + veryfast + faststart + a sane keyframe cadence is the volume-import workhorse.
  • AV1 saves 30–50% bitrate but multiplies encode cost — baseline H.264 everywhere, AV1 for top-traffic content only.
  • Watermarks and intros are burn-ins — decide them before bulk-processing a catalog, not after.
  • At 50k videos the ladder alone is ~26–56 TB — storage servers + CDN beat cloud object storage by a wide margin at tube scale.

Encoding is where a tube site’s economics are actually decided. Storage and bandwidth are the two bills that scale with success, and both are functions of one thing: how you encode. Yet search for tube-specific encoding guidance and you find generic codec-war articles written by cloud vendors whose conclusion is always “too hard, use our platform”. This guide is the opposite: the FFmpeg settings, rendition ladder and storage math for a self-hosted tube site in 2026, from someone on the self-hosted side of the argument.

Bitrates and sizes below are working defaults for typical 2D adult content, not laws. Codec licensing positions and browser-support percentages are as reported mid-2026 — verify before committing a 200k-video catalog to any single codec.

The rendition ladder: three rungs, maybe four

Streaming platforms publish ladders with eight rungs because they serve everything from smart TVs to elevators. A tube site needs three, sometimes four:

RenditionResolutionH.264 bitrate (≈)Size per 10 min (≈)Role
240p426×240300–500 kbps25–40 MBSlow mobile networks, instant start
480p854×4801.0–1.5 Mbps80–115 MBThe workhorse — most tube traffic plays here
720p1280×7202.5–3.5 Mbps190–265 MBDefault “HD”, the quality users expect free
1080p (optional)1920×10804.5–6 Mbps340–450 MBPremium/member tier, source-quality showcase

Two decisions matter more than the exact numbers. Whether to keep sources: re-encoding is only ever possible from what you kept, so keep originals for exclusive/licensed content and let feed content live as renditions only. Whether 1080p is free: making full HD a member perk is one of the oldest working tube monetization levers, and it cuts your bandwidth bill at the same time. (TubePress lets you define the format list itself: which renditions exist, with what FFmpeg options.)

FFmpeg settings that actually matter

A tube-grade H.264 encode is four decisions, then a template. Quality mode: use CRF for consistent perceptual quality (start at -crf 23, drop to 21–22 for premium tiers) with -maxrate/-bufsize caps so one high-motion scene cannot blow up delivery. Speed: -preset veryfast is the volume-import workhorse — the file-size penalty versus slow is real but modest, while the CPU-time difference is 5–10×; reserve slower presets for premium content. Seeking: force a keyframe cadence (-g 48–120) so scrubbing — which tube users do constantly — lands quickly. Delivery: -movflags +faststart is non-negotiable for progressive MP4, or the player waits for the whole file before starting. Put together, a 480p workhorse rendition looks like:

ffmpeg -i src.mp4 -vf scale=-2:480 -c:v libx264 -preset veryfast -crf 23 -maxrate 1.5M -bufsize 3M -g 96 -c:a aac -b:a 96k -movflags +faststart out_480.mp4

Beyond the renditions themselves, the encode pipeline also produces the assets that drive clicks: poster thumbnails, animated hover previews (short low-fps WebP/MP4 strips sampled across the timeline — disproportionately important for CTR), and, if you brand your content, watermarks. Watermarks and intro bumpers are burn-ins — a re-encode, not an overlay — so decide them before bulk-processing a catalog, not after. TubePress’s transcoding pipeline generates renditions, thumbnails and previews from an admin-defined format list, with watermark and intro burn-in options — the same FFmpeg logic above, wired into the upload and bulk-import flow.

AV1 and H.265 in 2026: the honest answer

The codec-war articles are right about one thing: AV1 delivers on the order of 30–50% bitrate savings over H.264 at comparable quality, and browser support is now broad on current versions. The part they skip is the operator’s cost table:

  • Encode cost explodes. Software AV1 encoding runs many times slower than x264 at tube-relevant settings. At catalog scale that is a real server bill — hardware AV1 encoders (recent GPUs and some CPUs) change this, if your encode boxes have them.
  • Old devices still play H.264. Adult traffic skews toward a very long device tail; a codec your tail cannot decode is a support ticket, so AV1 ships alongside H.264, not instead of it — which means more storage, not less, until you prune.
  • Licensing: AV1 is royalty-free by design (patent-pool claims against it have been reported over the years — the ecosystem has largely proceeded regardless); H.265’s multi-pool licensing mess is the main reason the web mostly skipped it for free content.

The pragmatic 2026 policy for a self-hosted tube: H.264 everywhere as baseline; add AV1 renditions for your top-traffic content only, where bandwidth savings repay the encode cost; ignore H.265 for web delivery unless your source workflow hands it to you. It is telling that the big adult CMS platforms only shipped AV1/H.265 support in their 2026 releases — the mainstream of this industry is exactly here.

The storage math, by catalog size

Using the 3-rung ladder (240+480+720, ≈ 0.35 GB per 10-minute video all-in) and the 4-rung ladder (≈ 0.75 GB with 1080p), average title length 15 minutes:

Catalog3-rung ladder4-rung ladder+ Sources kept (≈)
10,000 videos≈ 5 TB≈ 11 TB+ 8–15 TB
50,000 videos≈ 26 TB≈ 56 TB+ 40–75 TB
200,000 videos≈ 105 TB≈ 225 TB+ 160–300 TB

Rules that fall out of the table: dedicated storage servers with big HDDs beat cloud object storage on price at tube scale by a wide margin (do your own math against current egress fees — the difference is usually not close); renditions are re-creatable, so RAID redundancy plus a source backup beats backing up everything twice; and pruning the 1080p rendition of videos nobody watches is free money. The hosting guide covers the delivery side — CDN choices, what the big CDNs’ terms actually allow for adult, and reference builds by budget.

“Never self-host video” — the vendor claim, run through a calculator

The cloud-video industry publishes a steady stream of “15 reasons never to self-host video” content. The reasons are real for a five-person SaaS hosting product demos; they invert at tube scale. Hosted video platforms price by minutes stored and minutes delivered — at published mid-2026 rates, a modest 50k-video catalog with real traffic prices into five figures monthly, and that is before the platforms’ content policies meet your content (most mainstream video SaaS terms exclude adult outright). A storage server plus CDN running the ladder above does the same job for a few hundred dollars a month. Self-hosting video is not the hard mode of running a tube site; it is the business model.

TubePress turns this whole pipeline into configuration: admin-defined formats and renditions, FFmpeg transcoding with thumbnails, hover previews, watermark burn-in and multi-server storage — free and self-hosted. Download it and encode like it is your bill, because it is.

FAQ

Frequently asked questions.

Do I need 4K renditions on a tube site?
Almost never. Storage and bandwidth costs outrun real demand, and the device tail that watches free tube content rarely benefits. A 1080p premium tier is the practical ceiling for most operators.
Is AV1 worth it in 2026?
For your top-traffic content, with hardware encoders available — yes, the 30–50% bandwidth saving is real. As a wholesale H.264 replacement — not yet: encode cost and the old-device tail both argue for AV1 alongside H.264, not instead of it.
Should I encode on GPU or CPU?
CPU x264 wins on quality-per-bit and simplicity; hardware encoders (NVENC, QSV, and recent AV1 blocks) win when throughput is the bottleneck, at a small quality cost per bitrate. Volume importers usually end up hybrid: hardware for bulk, CPU for premium.
How much storage do 100,000 videos need?
At a 15-minute average: roughly 52 TB on the 240/480/720 ladder, about 112 TB adding 1080p — before keeping any sources. The math scales linearly, which is why rendition policy is a financial decision.
Can I re-encode my catalog later?
Only from sources you kept. Renditions-only catalogs are locked into their current quality forever, so decide source retention per content class on day one: keep originals for exclusive/licensed material, let feed content live as renditions.

Ready to launch your tube site?

TubePress is free, self-hosted and yours to keep — no license fees, no ionCube, no lock-in. Start with 150 free credits.

Self-hosted · full editable source · no ionCube