ReelSDK vs self-hosted FFmpeg: $8 vs $84 to $130 for 1,000 encodes
Self-hosted FFmpeg vs an API: about 0.8 cents per minute of video, or $130 a month for a box
Transcoding a minute of 1080p video to 720p H.264 costs about 0.8 cents on ReelSDK. On your own 4 vCPU machine the same encode costs about a tenth of a cent on AWS EC2 or DigitalOcean while the machine is busy, but the machine costs $130 or $84 a month either way, so the API stays cheaper below about 11,000 to 17,000 such minutes a month.
- ReelSDK ffmpeg, $0.02 per compute minuteabout $8
- AWS Lambda at 7,076 MB, with egressabout $4
- DigitalOcean 4 vCPU droplet, always on$84
- AWS EC2 c7i.xlarge, always on, with egressabout $130
List prices, us-east-1 for AWS. EC2, DigitalOcean and Lambda use the 22-second encode we measured on 4 vCPUs; ReelSDK bills its machine's whole time on the job, about 24 seconds. Each option includes downloading each 19 MB output once: AWS bills that at $0.09 per GB, the droplet includes 5,000 GiB of transfer and ReelSDK bills compute only.
Which one to pick
Pick ReelSDK if
- You transcode less than about 11,000 minutes a month, or in bursts: there's no idle machine, billing is by the second and failed jobs are free
- You don't want to build and run the queue, workers, storage, retries and ffmpeg updates around the one-line command
- You need it working today: the job below was one HTTP request and came back in about 26 seconds with a download URL
Pick self-hosted FFmpeg if
- You run tens of thousands of minutes a month at a steady rate, so a busy box costs about a tenth of a cent a minute against about 0.8 cents on ReelSDK
- Your jobs go past our limits: runs over 10 minutes, inputs over 500 MB, or options we block like two-pass encoding
- The video has to stay inside your own cloud account or network
Price per minute
Self-hosted FFmpeg and ReelSDK, per minute of video
The workload is one minute of 1920x1080 at 30 fps, transcoded to 720p H.264 at crf 23 with 128k AAC. ReelSDK bills raw ffmpeg at $0.02 per compute minute, by the second; see the pricing page. For your own machines we use the 22-second encode we measured on 4 vCPUs and today's list prices. The per-minute figures assume the machine is busy all the time; the last column is a quieter month.
| Option | List price | Per minute of video, busy | 1,000 minutes a month |
|---|---|---|---|
| AWS EC2 c7i.xlarge, 4 vCPU | $0.1785 an hour | 0.11 cents | About $130 always on, with egress |
| DigitalOcean CPU-Optimized, 4 vCPU | $0.125 an hour, $84 a month | 0.08 cents | $84 always on, transfer included |
| AWS Lambda, 7,076 MB | $0.0000166667 per GB-second | 0.25 cents | About $4 with egress |
| ReelSDK ffmpeg | $0.02 per compute minute | 0.79 cents | About $8 |
The per-minute column is compute only. AWS bills data transfer out at $0.09 per GB for the first 10 TB a month, about 0.16 cents each time a 19 MB output is downloaded. EC2 and DigitalOcean bill by the second with a 60-second minimum, so a queue that starts machines only when there's work moves them toward the busy figure, if you build that autoscaling. Prices from Amazon EC2 On-Demand pricing, checked November 2026.
Head to head
ReelSDK and self-hosted FFmpeg, feature by feature
| ReelSDK | self-hosted FFmpeg | |
|---|---|---|
| This job | About 0.8 cents; about 26 seconds from request to download URL | 0.08 to 0.25 cents of compute, plus storage and egress |
| Idle cost | $0 | $84 to $130 a month per 4 vCPU box; $0 on Lambda |
| Setup | An API key and one HTTP request | A machine or function, an ffmpeg build, a queue, storage and retries |
| Throughput | 3 running jobs and 500 a day per key, raised on request | About 165 of these encodes an hour per busy 4 vCPU box |
| Limits | 500 MB per input, 1 GB per job, 10 minutes per job | Your hardware; Lambda stops at 900 s and 10,240 MB |
| FFmpeg | ffmpeg 8.1, kept up to date by us | Any build you like, and you patch it |
| Failures | Failed jobs are free | You pay for the compute either way |
| Results | Signed URLs kept 7 days, or a webhook signed with HMAC-SHA256, with up to 6 attempts over about 7 hours | Whatever storage and callbacks you build |
Be honest
Where self-hosted FFmpeg is the better choice
Cheaper at steady volume
A 4 vCPU box that never sits idle runs this encode for about a tenth of a cent a minute, against about 0.8 cents on ReelSDK. Past about 11,000 to 17,000 minutes a month at an even rate, your own machines cost less, and on AWS Lambda this minute costs about a quarter of a cent at any volume, plus egress.
No limits but yours
Run jobs longer than 10 minutes, inputs bigger than 500 MB, your own FFmpeg build, and options we block, such as network inputs, the movie filter, the tee muxer or two-pass encoding.
Video stays in your account
If compliance says files never leave your cloud account or network, run FFmpeg there. ReelSDK fetches inputs from your URLs and keeps outputs on its storage for 7 days.
Migration
Moving a render from self-hosted FFmpeg
Keep the ffmpeg command and send it as args to POST /v1/ffmpeg; inputs become URLs and outputs come back as signed download links. The request below came back in about 26 seconds and cost about 0.8 cents. See the output. For a per-op comparison with another hosted API, see ReelSDK vs FFmpeg Micro.
self-hosted FFmpeg today
# One worker run: fetch, encode, store, sign. The queue, retries, scaling,
# ffmpeg updates and cleanup are yours to build and run.
curl -sSfL -o in.mp4 "https://api.reelsdk.com/f/lib/demo/landscape.mp4"
ffmpeg -stream_loop -1 -i in.mp4 -t 60 -vf scale=-2:720 \
-c:v libx264 -crf 23 -c:a aac -b:a 128k out.mp4
aws s3 cp out.mp4 "s3://your-bucket/out/$JOB_ID.mp4"
aws s3 presign "s3://your-bucket/out/$JOB_ID.mp4" --expires-in 604800
The same render on ReelSDK
curl "https://api.reelsdk.com/v1/ffmpeg?wait=60" \
-H "Authorization: Bearer $REELSDK_KEY" \
--json '{
"inputs": {
"in.mp4": "https://api.reelsdk.com/f/lib/demo/landscape.mp4"
},
"args": [
"-stream_loop", "-1", "-i", "in.mp4", "-t", "60", "-vf",
"scale=-2:720", "-c:v", "libx264", "-crf", "23", "-c:a", "aac", "-b:a",
"128k", "out.mp4"
],
"outputs": ["out.mp4"]
}'
# 200 OK after 25.9 s, trimmed
# {
# "job": {
# "id": "job_71mn9gov67nt0p70ocuj",
# "op": "ffmpeg",
# "status": "succeeded",
# "duration_s": 60,
# "outputs": { "out.mp4": "https://api.reelsdk.com/f/out/job_71mn9gov67nt0p70ocuj/out.mp4?s=..." }
# }
# }Field by field
| self-hosted FFmpeg | ReelSDK |
|---|---|
curl -o in.mp4 URL | inputs: { "in.mp4": URL } |
ffmpeg ... | args: the same arguments as a list |
aws s3 cp and presign | outputs: signed URLs kept 7 days |
| Queue and worker | ?wait=60 for short jobs, or a poll URL or webhook for long ones |
| Retry logic | An Idempotency-Key header; failed jobs are free |
| Autoscaling | Nothing; jobs run on fresh machines |
FAQ
ReelSDK vs self-hosted FFmpeg, answered
How much does it cost to run FFmpeg yourself per minute of video?
For a 1080p to 720p H.264 transcode, about a tenth of a cent per minute on an AWS c7i.xlarge or a DigitalOcean 4 vCPU droplet, if the machine is busy all the time, and about a quarter of a cent on AWS Lambda at 7,076 MB. Those use the 22 seconds of encoding we measured on 4 vCPUs. ReelSDK charged about 0.8 cents for the same minute.
When is self-hosting FFmpeg cheaper than an API?
When the machine stays busy. An always-on c7i.xlarge costs about $130 a month and a 4 vCPU droplet $84, which pays for about 17,000 or 11,000 of these transcodes on ReelSDK. Below that, or with bursty traffic, the API costs less than a box; above it, your own machines do, plus the time to run them. AWS Lambda bills only while it runs, about a quarter of a cent for this minute plus egress, if you build the queue, storage and retries around it.
Can I run FFmpeg on AWS Lambda?
Yes, with a static ffmpeg binary in a layer or container image. Lambda gives up to 10,240 MB of memory, one vCPU per 1,769 MB, 10,240 MB of /tmp and 900 seconds per run. Our minute of video would cost about a quarter of a cent there, plus $0.09 per GB to download outputs from S3.
What can't I do with ReelSDK's FFmpeg API?
Jobs stop at 10 minutes, inputs at 500 MB each and 1 GB per job. Arguments can't contain URLs, absolute paths or parent-directory paths, and movie, amovie, the tee muxer and two-pass encoding are blocked. Default limits are 3 running jobs and 500 jobs a day per key.
Other comparisons
Sources, checked November 2026
Ready to build?
Ship video from your app today
Your first $0.50 is free, and one key covers every endpoint.
