video-encodingsocial-mediaffmpeg

Compress video for TikTok at the cap, not below. It re-encodes.

··10 min read
Compress video for TikTok at the cap, not below. It re-encodes.

Your export settings panel says 4,000 kbps and your Reel still looks soft. You drop it to 2,500 and it looks the same. You push it to 20,000 and it looks the same again, just takes four minutes to upload.

To compress video for TikTok and Instagram Reels, target 1080x1920 H.264 High profile at 8 to 10 Mbps with AAC-LC audio at 128 kbps, and stop there. Both platforms re-encode every upload, so bits you spend below their ingest ceiling get discarded by their transcoder and bits above it buy upload time only. Resolution and aspect ratio survive their pass. Your bitrate does not.

What TikTok and Instagram actually deliver to a viewer

The stream that reaches a viewer's phone is the platform's own re-encode of your file, and it runs at a far lower bitrate than any export preset suggests.

An October 2026 measurement of 135 short videos across the three platforms, published as short-form-video-delivery-2026, found the median best available stream ran 2.1 Mbit/s on Instagram Reels, 1.1 Mbit/s on TikTok, and 5.7 Mbit/s on YouTube Shorts. The same measurement found TikTok never offered better than 1080p, Reels offered 1440p on only 4% of videos, and Shorts offered 2160p on 19%.

Meta's engineering team described the mechanism in its post on how Facebook encodes your videos: an uploaded video gets encoded into a ladder of resolutions at multiple CRF values, with higher-quality encodes applied later to videos that accumulate watch time. Your upload is an input to that ladder. It is never what anyone plays.

Meta has since moved most of that delivery to AV1. Its AV1 for Reels post reports AV1 passing 70% of Reels watch time on iPhone and about a 12% average bitrate reduction from the switch. So the codec you upload is not even the codec that gets played.

One target per destination

Pick the ceiling each platform publishes, land just under it, and move on. These are upload targets, chosen so the platform's scaler and tone-mapper have nothing to do.

DestinationResolutionVideo bitrateCodecAudioPublished source
Instagram Reels1080x19208-10 Mbps (hard cap 25)H.264 High, 30 fpsAAC-LC 128 kbps[Meta's IG User media reference](https://developers.facebook.com/docs/instagram-platform/instagram-graph-api/reference/ig-user/media)
TikTok1080x19208-10 MbpsH.264, 30 fpsAAC-LC 128 kbps[TikTok Content Posting API guide](https://developers.tiktok.com/docs/en/content-posting-api-media-transfer-guide)
YouTube Shorts1080x19208 Mbps at 30 fps, 12 at 60H.264 High, closed GOP, 4:2:0AAC-LC 128 kbps[YouTube recommended upload encoding](https://support.google.com/youtube/answer/1722171?hl=en)

Meta's publishing reference caps Reels at VBR up to 25 Mbps, 300 MB, 23 to 60 fps, and recommends 9:16 explicitly to avoid cropping. TikTok's API guide accepts 360 to 4096 px per side at 23 to 60 fps, up to 4 GB and 10 minutes, with H.264 recommended and chunked upload above 64 MB. Nobody needs the ceiling. 8 Mbps at 1080x1920 is a clean source by any of these specs and uploads far faster than a 25 Mbps file.

The FFmpeg command for all three is the same:

ffmpeg -i master.mp4 \
  -vf "scale=1080:1920:force_original_aspect_ratio=decrease,pad=1080:1920:(ow-iw)/2:(oh-ih)/2,format=yuv420p" \
  -c:v libx264 -profile:v high -b:v 8M -maxrate 10M -bufsize 16M \
  -r 30 -g 60 -keyint_min 60 -sc_threshold 0 \
  -colorspace bt709 -color_primaries bt709 -color_trc bt709 \
  -c:a aac -b:a 128k -ar 48000 -ac 2 \
  -movflags +faststart out.mp4

If the clip is already 9:16 at the right height, the scale filter is a no-op.

The asymmetry: rate converges, geometry does not

A 50 Mbps master and an 8 Mbps master land in the same place after Instagram's pass. A 1080x1920 file and a 3840x2160 file do not.

When you hand Instagram a 4K vertical file, its transcoder downscales it to feed the ladder. That rescale is a generation of resampling you paid for with nothing, and it happens with whatever scaler their pipeline uses rather than the one you'd pick. When you hand it exactly 1080x1920, the top rung of the ladder is a straight re-encode at native geometry. Same for a 1440x1920 file with a 4:3-ish ratio: Meta's own spec says 9:16 avoids cropping, which means anything else gets cropped or padded, and a pad is black bars baked into the delivered pixels forever.

The 4K half of this is written up separately in your 4K AI video isn't TikTok-ready, convert it to 1080x1920: downscale before upload, every time.

Colour signalling is the other thing that survives, and it fails loudly. Meta documented its own version of this in the HDR video on Reels post: washed-out colours on Meta's Android HDR path on newer devices, until the team built custom PQ and HLG to SDR tone mapping. Meta now tone-maps server-side from the original HDR.

If Meta's engineers had to build that, an untagged 10-bit render from a generative model has no chance of being read correctly. That's why a Kling or Veo output can come back grey with a perfectly healthy 20 Mbps bitrate. Kling 4.0 washed out on TikTok? Fix the tag, then the 9:16 covers the tone-map in full.

Audio is a third thing you don't need to fight. The same October 2026 measurement found Instagram normalises Reels audio to roughly -14 LUFS, with 92% of measured Reels within ±1 LU and none louder than -14.0. Hand it AAC-LC at 128 kbps and let their loudness pass do the rest.

Which settings are provably wasted effort

Every knob below changes your file and changes nothing about what a viewer sees, because the platform's own transcoder overwrites the decision.

  • Two-pass encoding. Two-pass buys you precise rate control at a target size. The platform re-derives its own rate from scratch, so your precision dies on their first frame. It also doubles your encode time, and on a stateless job API you'd have to carry a pass-1 stats file between runs.
  • CRF 18 or anything below CRF 20. A visually lossless master at CRF 18 and a clean CRF 21 encode converge after a re-encode down to 2 Mbps. The first one just costs more upload.
  • Anything above 10 Mbps. Reels caps ingest at 25 Mbps and delivers around 2. You can give it 24 if you enjoy waiting.
  • 60 fps for talking-head or AI-generated content. 60 fps doubles your frame count and halves the bits per frame inside the platform's own rate budget. For motion-light footage, 30 fps at the same rate looks better after their pass.
  • H.265 / HEVC ingest. HEVC compresses better, which is irrelevant when the file is decoded and thrown away. Meta accepts it; TikTok recommends H.264. HEVC also drags in 10-bit and HDR tagging problems, which is the one failure mode that does reach the viewer. Upload H.264 8-bit.
  • 4K anything. TikTok never served better than 1080p in the measured sample. A 4K upload buys a rescale you didn't ask for.
  • Compressing to 30-50 MB "for faster upload." Faster upload is a real benefit and a real cost: those bits are gone from the ladder's top rung. If upload speed is the constraint, cut the bitrate to 8 Mbps and keep the resolution. Never trade resolution for upload time.

The one caveat: TikTok's in-app upload caps (widely repeated as about 287.6 MB on iOS and 72 MB on Android) are not published on an official TikTok page, so treat them as folklore. Publishing through the Content Posting API gives you the documented 4 GB ceiling and takes the question off the table.

Doing this to 200 clips

A single export is a slider. Two hundred exports a week is a pipeline, and that's where the manual route breaks down rather than where the settings do.

Running that FFmpeg command yourself costs nothing in software and requires a machine with FFmpeg on it, a queue, and someone watching when a job hangs on a corrupt input. In n8n, Make, or Zapier there's no FFmpeg binary at all, so the command above isn't available to you. The same normalize-to-9:16-at-8-Mbps job is one call to FFmpeg Micro from code, from an n8n or Make node, or from an AI agent over MCP, with no encoder to host and no timeout to babysit. Billing is per second of input video, so a 30-second clip costs 30 seconds whether you compress it hard or barely at all. Failed jobs are never billed, which matters when a batch of 200 contains three broken files. The free tier gives you 200 tokens, about 33 minutes of video, with no card.

If the job is specifically the geometry (reframe to 9:16, 1:1, 4:5 or 16:9, pad or crop), the Resize for Format blueprint is that exact run as a form: upload, pick the ratio, download.

Verify before you publish, whichever route you take. Read the output back with ffprobe and check three fields:

ffprobe -v error -select_streams v:0 \
  -show_entries stream=width,height,bit_rate,color_primaries,pix_fmt \
  -of default=noprint_wrappers=1 out.mp4

You want width=1080, height=1920, color_primaries=bt709, pix_fmt=yuv420p. If color_primaries comes back unknown, the tag didn't stick and that is the single most likely cause of a washed-out upload. Instagram's publishing container rejects on encode-level causes with codes like 2207052, so catching a bad file with ffprobe before it reaches the Graph API saves a round of guessing.

When the destination is a hard byte cap rather than a re-encoding platform, how to compress video: five reasons, five different settings sorts the cases. If the master is a long-form video you're cutting clips from, Instagram Features long form video: 8 Reels from one master covers the split.

FAQ about compressing video for TikTok and Reels

What is the best bitrate for Instagram Reels?

The best bitrate for Instagram Reels is 8 to 10 Mbps at 1080x1920 with H.264 High profile. Meta's publishing spec allows VBR up to 25 Mbps, but delivered Reels streams measured at a median 2.1 Mbit/s in October 2026, so anything above 10 Mbps only lengthens your upload.

Does compressing a video harder make processing cheaper?

Compressing harder does not make FFmpeg Micro processing cheaper. Billing is per second of input video duration, so a 60-second clip costs 60 seconds of metering whether it arrives as a 10 MB file or a 400 MB one. Trimming the duration is what lowers the cost.

Should I upload H.265/HEVC to TikTok since it compresses better?

Upload H.264, not HEVC, to TikTok. HEVC's better compression gains you nothing on a file that gets decoded and re-encoded on arrival, TikTok's own Content Posting API guide recommends H.264, and HEVC commonly arrives with 10-bit or HDR tagging that produces washed-out colour after the platform's SDR conversion.

Why is my Reel blurry when the bitrate was high?

A blurry Reel at high bitrate is almost always geometry or colour, not rate. The usual causes are a source that isn't exactly 1080x1920 (so Instagram's scaler resamples it), an aspect ratio other than 9:16 (so the frame gets cropped or padded), or a missing Rec.709 tag on a 10-bit render.

Can I compress video for TikTok inside n8n or Make?

You can compress video for TikTok inside n8n and Make by calling a video API, since neither tool ships an FFmpeg binary. FFmpeg Micro has an n8n community node and a Make app, and the job is submit, poll for status, download the finished MP4.

Anyone publishing the same vertical spec on a schedule, 20 clips a week or 200, is really running one encode decision over and over: 1080x1920, H.264, 8 Mbps, bt709 tagged, verified with ffprobe. The patterns behind automating that across TikTok, Reels and Shorts are collected on the social media automation page, and Sign up free gets you running it today: no servers to run, called from your code, from n8n, Make and Zapier, or from your AI agents.

About Javid Jamae

Founder & CEO at FFmpeg Micro

Javid is a software engineer, author, and entrepreneur with over 25 years of professional software development experience across enterprise, startup, and consulting environments. He founded FFmpeg Micro to make video processing accessible to developers through a simple, automation-first REST API.

Software EngineeringVideo ProcessingFFmpegCloud ArchitectureAPI DesignAutomation

Ready to process videos at scale?

Start using FFmpeg Micro's simple API today. No infrastructure required.

Get Started Free