Discover how we embedded FFmpeg inside a browser agent to process videos instantly, no servers, APIs, or uploads. Watch how it all comes together.
Read the original article
Comments
An aside but related?
FFmpeg has complex syntax because it’s dealing with the _complexity of video_. I agree with everyone about knowing (and helping create or contribute to) our tools.
Today I largely forget about the _legacy_ of video, the technical challenges, and how critical it was to get it right.
There are an incredible number of output formats and considerations for _current_ screens (desktop, tablet, mobile, tv, etc…). Then we have a whole other world on the creation side for capture, edit, live broadcast…
On legacy formats it used to be so complex with standards, requirements, and evolving formats. Today, we don’t even think about why we have 29.97fps around? Interlacing?
We have a mix of so many incredible (and sometimes frustrating) codecs, needs and final outputs, so it’s really amazing the power we have with a tool like FFmpeg… It’s daunting but really well thought out.
So just a big thanks to the FFmpeg team for all their incredible work over the years…
By echelon 2025-11-0416:45 > FFmpeg has complex syntax because it’s dealing with the _complexity of video_.
It's dealing with 3D data (more if you count audio or other tracks) and multi-dimensional transforms from a command line.
By shardullavekar 2025-11-0415:53 no 2nd thoughts about it, we are only making ffmpeg more accessible and embeddable.
By charcircuit 2025-11-0417:281 reply >FFmpeg has complex syntax because it’s dealing with the _complexity of video_
It's complexity paired with bad design, making the situation worse than it could be.
By SpaceManNabs 2025-11-0422:23 I refuse to admit that ffmpeg is bad design until I see a better one. so if you have a better one I am all ears because it would surely be very illuminating.
By hexo 2025-11-051:03 what legacy formats? what are you even talking about?
Have to admit, ffmpeg syntax is not trivial... but also the project is 24 years old and is basically the defacto industry standard. If you believe you will still be editing videos in 20 years with the CLI (or any other tool or any programming language) wrapping it then it's probably worth few hours learning how it actually works.
The syntax isn't too bad. The problem is that I have to use it a couple of times a year, on average. So every time I've forgotten and have to relearn. This doesn't happen with GUIs nearly as much, and it's why I prefer them over CLI tools for anything that I don't do at least once every week or two.
That’s why you write scripts, or put a couple snippets in your notes.
I do have snippets in my notes. The problem is that nearly every time I use it, I need to do something different than the previous time.
By bigiain 2025-11-051:41 I started making sure my notes included the search queries keywords/keyphrases/terminology I uses and any webpages or documentation that I successfully used to come up with the solution (and where relevant, and search queries or misunderstandings on my part that lead to dead ends).
This hasn't solved the problem of sometimes needing to do new things, but it at least gives me a map to remind me of the parts of the rabbithole I've explored before.
My question/curiosity is why do so many people use ffmpeg (frustrated by the syntax) when GStreamer is available?
`gst-launch-1.0 filesrc ! qt4demux ! matroskamux ! filesink...` people would be less frustrated maybe?
People would also learn a little more and be less frustrated when conversation about container/codec/colorspace etc... come up. Each have a dedicated element and you can better understand its I/O
By throwaway2046 2025-11-0419:05 I haven't tried GStreamer but I found FFmpeg to be extremely easy to compile as both a command line tool and library, plus it can do so much out of the box even without external libraries being present. It's already used in pretty much everything and does the job so it never occurred to me (or others) to look for alternatives.
By thisislife2 2025-11-0512:14 Handbrake ( https://handbrake.fr/ ) and AviDemux ( https://www.avidemux.org/ ) is what the average user needs. Subler ( https://subler.org/ ) is also a good macOS app for muxing and tagging.
By goeiedaggoeie 2025-11-0414:50 Still has a way to go, but very exciting.
I agree, I suggest using this instead : https://github.com/kkroening/ffmpeg-python . While not perfect once you figure it out it is far easier to use and you can wrap more complicated workflows and reuse them later.
Kkroening's wrapper has been inactive for some time. I suggest using https://github.com/jonghwanhyeon/python-ffmpeg instead. It has proper async support and a better API.
By artpar 2025-11-0413:40 I think that goes with almost every tool you want to use with llm. User should already know the tool ideally so mistakes by llm can be prevented before they happen.
Here making ffmpeg as "just another capability" allows it to be stitched together in workflows
By shardullavekar 2025-11-0413:401 reply true, companies like Descript, Veed, or Kapwing exist because no coders find this syntax intimidating. Plus, a CLI tool stands out of a workflow. We wanted to change that.
Don't "no coders" find the concepts described in this article imdimitating?
The article states that whatever the article is trying to describe "Takes about ~20-30 mins. The cognitive load is high....". while their literal actual step of "Googling "ffmpeg combine static image and audio."" gives you the literal command you need to run from a known source (superuser.com sourced from ffmpeg wiki).
Anyone even slightly familiar with ffmpeg should be able to produce the same result in minutes. For someone who doesn't understand what ffmpeg is the article means absolutely nothing. How does a "no coder" understand what a "agent in a sandboxed container" is?
By shardullavekar 2025-11-0414:101 reply we took a basic example and described it. (will try adding a complex one)
we have our designer/intern in our minds who creates shorts, adds subtiles, crops them,and merges the audio generated. He is aware of ffmpeg and prefers using a SaaS UI on top of it.
However, we see him hanging out on chatgpt, or gemini all the time. He is literally the no coder we have in mind.
We just combined his type what you want + ffmpeg workflows.
Wouldn't that intern just use an NLE (be it Premiere, Davinci Resole etc) anyway? If you need to style subtitles and edit shorts and video content, you'll need a proper editor anyway.
By shardullavekar 2025-11-0415:38 1. download a larger video from s3. 2. Use NLE and cut it into shorts. (crop, resize, subtitles etc.) 3. Upload shorts on YouTube, Instagram, Tiktok.
He does use davinci resolve but only for 2.
NLEs make ffmpeg a standalone yet easy to use tool.
Not denying that major heavy lifting is done by the NLE. We go a step ahead and make it embeddable in a larger workflow.
By javier2 2025-11-0420:55 ffmpeg is pretty complicated, but at least it actually works.
By somat 2025-11-0421:00 The thing that helped me get over that ffmpeg bump, where you go from copying stack overflow answers to actually sort of understanding what you are doing is the fairly recent include external file syntax. On the surface it is such a minor thing, but mentally it let me turn what was a confusing mess into a programing language. There are a couple ways to evoke it but the one I used was to load the whole file as an arg. Note the slash, it is important "-/filter_complex filter_file"
https://ffmpeg.org/ffmpeg-filters.html#toc-Filtergraph-synta...
"A special syntax implemented in the ffmpeg CLI tool allows loading option values from files. This is done be prepending a slash ’/’ to the option name, then the supplied value is interpreted as a path from which the actual value is loaded."
For how critical that was to getting over my ffmpeg hump, I wish it was not buried halfway through the documentation, but also, I don't know where else it would go.
And just because I am very proud of my accomplishment here is the ffmpeg side of my project, motion detection using mainly ffmpeg, there is some python glue logic to watch stdout for the events but all the tricky bits are internal to ffmpeg.
The filter(comments are added for audience understanding):
and the ffmpeg evocation:[0:v] split #split the camera feed into two parts, passthrough and motion [vis], scale= #scale the motion feed way down, less cpu and it works better w=iw/4: h=-1, format= #needed because blend did not work as expected with yuv gbrp, tmix= #temporial blur to reduce artifacts frames=2, [1:v] #the mask frame blend= #mask the motion feed all_mode=darken, tblend= #motion detect actual, the difference from the last frame all_mode=difference, boxblur= #blur the hell out of it to increase the number of motion pixels lr=20, maskfun= #mask it to black and white low=3: high=3, negate, #make the motion pixels black blackframe= #puts events on stdout when too many black pixels are found amount=1 [motion]; #motion output [vis] tpad= #delay pass through so you get the start of the event when notified start=30 [original]; #passthrough outputff_args = [ 'ffmpeg', '-nostats', '-an', '-i', camera_loc, #a security camera '-i', 'zone_all.png', # mask as to which parts are relavent for motion detection '-/filter_complex', 'motion_display.filter', #the filter doing all the work '-map', #sort out the outputs from the filter '[original]', '-f', 'mpegts', #I feel a little weied using mpegts but it was the best "streaming" of all the formats I tried 'udp://127.0.0.1:8888', #collect the full video from here '-map', '[motion]', '-f', 'mpegts', 'udp:127.0.0.1:8889', #collect the motion output from here, mainly for debugging ]By sexyman48 2025-11-0414:29 [dead]
As someone who has used ffmpeg for 10+ years maintaining a relatively complex backend service that's basically a JSON to ffmpeg translator I did not fully understand this article.
Like the Before vs after section doesn't even seem to create the same thing, the before has no speedup, the after does.
In the end it seems they basically created a few services ("recipes") that they can reuse to do simple stuff like speed-up 2x or combine audio / video or whatever
By shardullavekar 2025-11-0413:441 reply thanks for calling it out, I will correct the before vs after section. But you can describe any ffmpeg capability in plain English and the underlying ffmpeg tool call takes care of it.
I have written a lot of ffmpeg-python and plain ffmpeg commands using LLMs and while I am amazed at how good Gemini or chatGPT can handle ffmpeg prompts it is still not 100% so this seems to me like a big gamble on your part. However it might work for most users that only ask for simple things.
By shardullavekar 2025-11-0413:561 reply so creators on 100x will create well defined workflows that others can reuse. If a workflow is not found, llm creates one on the go and saves it.
That sounds good, save the LLM generated workflows and have them edited by more seasoned users.
Or you could go one step further and create a special workflow which would allow you to define some inputs and iterate with an LLM until the user gets what he wants but for this you would need to generate outputs and have the user validate what the LLM has created before finally saving the recipe.
By shardullavekar 2025-11-0414:59 That's exactly how it is implemented!
