top | item 44520855

Show HN: FFmpeg in plain English – LLM-assisted FFmpeg in the browser

179 points| bjano | 7 months ago |vidmix.app

I found that I am using ChatGPT more and more to get the FFmpeg command I need, but the process can be a bit tedious: copy-pasting commands, dealing with input file names and locations, making sure the prompt contains enough info about the input files.

This site attempts to solve that. You just describe what you want to do, pick the input files and an LLM (currently DeepSeek) generates the FFmpeg command. You can then run it directly in your browser or use the command elsewhere.

46 comments

order

jjbinx007|7 months ago

I recently used Gemini to help with some dashcam videos that weren't being saved properly. I was sure most of the data were there but the files wouldn't play in VLC or MPC, so I asked Gemini.

It suggested various things to try, and after pasting in the error messages each time it suggested more and more radical things. Eventually it suggested a program called Untrunc, where you give it a working video file as a reference and then the file that's faulty and as if by magic it worked!

Just wanted to mention this in case anyone else is struggling to get FFmpeg to repair a file.

pjc50|7 months ago

Interesting - I took a look at how this works and why it needs a reference, and the answer is the usual one of the 'moov atom': a critical piece of metadata. Lots of programs output it at the end of the file, but that makes it vulnerable to truncation, and for streaming it's useful to move it to the beginning of the file. ffmpeg refers to this as "fast start".

kkukshtel|7 months ago

To yes, and this —

I've started doing something similar on the command line with Claude Code that works incredibly well:

claude -p "use ffmpeg to convert myvideo.mov into an h264 video suitable for youtube upload" --dangerously-skip-permissions

Highly recommended! I use --dangerously-skip-permissions because I just want to set it and forget it and dont need to babysit the run.

blueboo|7 months ago

`llm cmd` by simonw is even better for one-off commands. It spits out the invocation and you can repurpose as suits your needs. Thanks Simon!

teddyh|7 months ago

Em-dash alert.

jjcm|7 months ago

Great work on this - I made a terminal command similar to this (llmpeg), and was actively trying to get exactly this working - a webasm compiled version of ffmpeg that could encode in the browser. I for the life of me couldn't get the provided examples on https://github.com/ffmpegwasm/ffmpeg.wasm to run.

Just for my own development curiosity, was there anything specific you had to do to get ffmpegwasm to work?

bjano|7 months ago

The creator of that repo has a blog with some posts on building a wasm version of FFmpeg, I mostly followed those: https://jeromewus.medium.com/ I also used some of their scripts for building the 3rd-party libraries.

It took a bit of trial and error to see which versions of the different libraries and build tools work together.

iamflimflam1|7 months ago

I had the same problem - would love to see a working example.

toddmorey|7 months ago

The Warp terminal[1] is excellent for this type of thing. In agent mode, you just describe what you want to have happen and it generates the proper command(s) (that you can approve before running).

I use it a lot to convert videos and turn a folder of tiff files into pngs at 1/2 size, etc. It's great at generating FFMEG commands and chaining the right tools together.

[1] https://www.warp.dev

WesleyLivesay|7 months ago

I was going to mention Warp here as well. It is fantastic when it comes to almost anything in the terminal. It has caused me to use the terminal a lot more on all of my computers because I don't have to spend a bunch of time poking around on Google to find the command to run.

I have used it for ffmpeg and then a lot of other slightly more complex commands. A recent one from the other day was gathering up all of the .epub documents in a directory tree, renaming them to the name of the directory they were in, and then placing them all in one single directory. That would have been a whole project for me, and Warp gave me the command with just that description. Any LLM interface would have done the same, but Warp just let me hit "Enter" and run it, no need to copy and paste.

d4rkp4ttern|7 months ago

Also in zed you can hit ctl-enter in terminal or in code and get the same

orderone_ai|7 months ago

This is such a convenient tool for a casual user, and a great application of an LLM to a narrow task that probably couldn't be handled quite so easily everywhere. Also a great example of the emerging 'chat driven' UX trend, which I'm really liking.

madeofpalk|7 months ago

I think it's amusing how the most consistent usecase for "LLM for CLI" is to use ffmpeg.

adamgusky|7 months ago

Anyone that has built something with FFmpeg understands why this is at the top of HN.

danielvaughn|7 months ago

I tried to implement something very similar recently, and had the hardest time getting the LLM to produce anything remotely resembling actual ffmpeg commands.

ilaksh|7 months ago

You were using a weak LLM then. The difference between one of the leading edge LLMs like Gemini 2.5 Pro, o3, or Claude 4 and an average LLM or one you can run on your typical PC/laptop is night and day.

victorbjorklund|7 months ago

Which model did you use? Claude has been spot on for all my needs.

cchance|7 months ago

load up the man pages into the context maybe?

jeanlucas|7 months ago

for real? to me it works fine with just chatGPT (free)

bsenftner|7 months ago

I think you need to add some liberal filename handling. I have directories of videos generated by various AI video models, and they have spaces in the filenames, not just one, but the prompt to generate the video plus the major parameters are all the filename. They are long, pains in the ass to work with, but they are there. Would be nice if your tool could work with them.

alalonde|7 months ago

Nice! Reminds me of the ooo-wow early LLM days where those little explain-Regexps one-off sites were popping up everywhere. Seems inevitable that the LLMs will slurp up this functionality natively as well

matcha-video|7 months ago

Love this! Folks here should also check out https://wide.video if they're looking for a VERY full featured free in-browser video editor

pdyc|7 months ago

interesting, i created a video editor to generate ffmpeg commands because it was difficult to adjust ffmpeg commands on commandline https://newbeelearn.com/tools/videoeditor/ . This lets you tweak ffmpeg commands visually.

guluarte|7 months ago

I also recommend djcopley/ShellOracle if you are lazy like me and never remember the commands.

raincole|7 months ago

That's great. How could it work in browser? Is it a webassembly implementation of FFmpeg?

p0w3n3d|7 months ago

  I'm sorry Dave, I'm afraid I cannot do that

runeks|7 months ago

How do you know my name?!

baobabKoodaa|7 months ago

This is a perfect use case for thin LLM wrappers!

yakattak|7 months ago

Honestly what a great application of LLMs. FFmpeg is a very powerful tool, and as with most powerful tools is very complicated to run correctly. Do the files get uploaded though? Or does it just grab the location on disk?

bjano|7 months ago

Nothing is uploaded, everything runs locally.

26d0|7 months ago

This appears to use ffmpeg.wasm