Create Vintage Videos Using FFmpeg in 4 Simple Steps | Hacker News Reader

Create Vintage Videos Using FFmpeg in 4 Simple Steps

ottverse.com

293 pointsponderingfish48 comments

Every single vintage effect misses out on one crucial factor - Resolution. Vintage films (even when scanned in 4K) do not have the same type of aliasing and sharpening aspects that digital cameras have. This effect is what “makes” it vintage for me.

If you haven’t seen, Apollo (2019) documentary is simply stunning and even though it’s in 4K, it appears vintage.

I'm curious what you're referring to specifically, since there is incredibly sharp film footage out there.

Are you talking about resolution from film grain? Because 35mm film is generally to be considered a good bit higher resolution than 4K.

Or the noise properties of film grain, which is a separate issue from resolution? Which shows up some in well-lit conditions, but is even more obvious in nighttime shots? That's certainly easy to add.

Or are you talking about lower-quality lenses (e.g. chromatic abberation), or focus issues?

Or something else? I don't know at all what you mean by the "aliasing and sharpening" that you claim professional digital cameras perform. (Since we're not talking about iPhone camera processing, but rather the cameras movies and broadcasters use.)

I've done A-list visual effects work since the 90s, and I can say that film for motion picture production[1] varies widely in it's grain density and size. This variation is amplified by the fact that DPs often "push" or "pull" the stock which means they treat it as if it was a different ISO speed than labeled (exposure and lab processing change to compensate). In practical application digital elements nearly always have to be reduced in effective resolution to match film plates when composited.

Of course under laboratory conditions using larger negative size (like 65mm), Velvia or some other extraordinary film stock you can of course produce extremely high resolution individual frames, but as we know from a lifetime of movies this is neither practical nor necessary. Though a great source of religious wars.

However, there is a reason why film's "lower resolution" still produces superior resolving power...

Every single frame of film has a different randomized set of sample points.

This in combination with human persistence of vision[2] and our ability to see pattern but discard noise means that a static shot on film is significantly more resolved in our minds after only a few frames than the equivalent shot recorded by color filter array[3]. It's not just perceptual, temporal sampling truly produces a more accurate representation of the image content over time.

This is a fascinating quality of film that can give you the sense of graininess and low quality while still producing superior image detail in your mind, but as soon as you stop the projector and evaluate the film image by image the effect is lost. You might be left wondering where did all that detail go?

[1]: https://en.wikipedia.org/wiki/List_of_motion_picture_film_st...

[2]: https://en.wikipedia.org/wiki/Persistence_of_vision

[3]: https://en.wikipedia.org/wiki/Color_filter_array

Analog support doesn't sample individual pixels in a grid format. That's why it's rather meaningless to talk about the resolution of analog photos or video, the actually information density will vary depending on a large number of factors and may not even be the same depending on what you measure. For instance old school analog TV had significantly more information in the lumas than in the chroma.

And then you have old school lenses vs. modern lenses and many other factors.

I agree with the parent that the end result is an amusing novelty but if the intention was to fool me into thinking that it was actually an ancient recording I don't think I would've fallen for it. It's too sharp, the blurring effect on movement feel wrong too and the dynamics are off I think (film has amazing contrast that digital sensors still struggle to emulate).

> For instance old school analog TV had significantly more information in the lumas than in the chroma.

That’s the case for the large majority of digital content as well, in a codec like H264 the chroma resolution is a quarter of the luma resolution.

> but if the intention was to fool me into thinking that it was actually an ancient recording

To be fair, no one claimed this. I think it’s almost impossible to do because so many factors come together.

Sure, but I wouldn't go so far as to say resolution of analog is "meaningless". I'm not aware of a precise quantitative measure of film "resolution", but simply from a qualitative sense of resolving detail, it's quite meaningful and useful to say that 35mm film is generally considered to lie somewhere between 4K and 8K. Which is far higher than 1080p.

In other words, if you're watching a 1080p video and complaining it appears too "high-resolution" to have been transferred from film, that's almost certainly incorrect, and it's something else that cluing you in it wasn't film -- probably grain and/or color characteristics.

Isn't the opposite the case? If I see aliasing and sharpening, I see it as digital distortion you wouldn't expect from film. When I see a bunch of digital artifacts on a film transfer, I think of it as a bad transfer.

If it's overly sharp, yes.

But most commercial video footage isn't digitally sharpened, for a host of reasons (mainly that it usually looks bad, harms more than it helps).

So high resolution doesn't mean sharpening has been applied. If it came from film, it just means the film had that level of detail to begin with.

For home movies, at least, 8mm and Super 8mm was a very common format. An 8mm frame has a lot less resolution than 35mm. Kodachrome and Ektachrome film have very identifiable artifacts, too.

Sorry, I am not well educated in this area to precisely define the “resolution”. Perhaps it’s film grain or the smoothness of the edges.

There's vintage and vintage. Old film scanned today (with modern resolution and chromatic fidelity, reproducing accurately chemical and optical issues in the original) is very different from old TV footage (with anisotropic filtering and old magnetic tape degradation effects) and from early digital cameras.

somewhat related:

https://www.gamasutra.com/blogs/KylePittman/20150420/241442/...

Then add that affect lol. In production you don't just apply one filter and be done with it, and you don't just apply the filter evenly, you want to layer mask and add something dynamic to it. A proper tutorial won't tell you about all the other filters, just the one specific effect it is talking about.

Excellent point. Perhaps, a blurring filter would help with that.

I would consider also getting some old lenses like the Krasnogorsk if you're recording this yourself.

https://youtu.be/2YzFPruwXyk

https://youtu.be/XelO_3rrkeA

Nice! I am actually just using my iPhone videos and converting them this way! The output is pretty good and when I shared them with family, they loved it. Thanks for the lens links.

And a little bit of grain?

Which Apollo documentary are you referring to? There were a few last year according to IMDB

This one. It's quite incredible in the way it's constructed.

https://en.wikipedia.org/wiki/Apollo_11_(2019_film)

> The film consists solely of archival footage, including 70 mm film previously unreleased to the public, and does not feature narration, interviews or modern recreations.

> In late 2016, Todd Douglas Miller had recently completed work on The Last Steps, a documentary about Apollo 17, when British archivist and film editor Stephen Slater suggested making a similarly themed documentary for the upcoming 50th anniversary celebrations of Apollo 11.

> Miller's conception of the film was centered on a direct cinema approach. The final film contains no voice-over narration or interviews beyond what was available in the contemporary source material.

Found a preview of this on youtube for anyone curious: https://www.youtube.com/watch?v=NgUYurzK-tM

The example footage looks awesome - incredible quality but unmistakably vintage.

Yes indeed, this one. It gives me chills what humans did back in the 1960's. The quality of the film amplifies this emotion by making it seem like it was just a normal everyday life in modern fidelity.

That clip may be max 1080. Whats crazy is that while the film was released in 4k, the content sourced from 70mm film has a way higher (8k? more?) potential resolution. Combined with something like https://news.ycombinator.com/item?id=25105713 and we may still have really cool things to see from the 60s.

Not OP but can only assume it's this one https://www.imdb.com/title/tt8760684/?ref_=fn_al_tt_1 lots of new footage and a great documentary.

For working with FFmpeg for programmatic video editing, I also recommend using moviepy: https://github.com/Zulko/moviepy

Discussion a few years ago: https://news.ycombinator.com/item?id=16297295

It’s cool to see more people writing posts and experimenting on this topic. I was surprised at how little info I could find trying to research this about a year ago, and spent a decent chunk of this year experimenting with trying to create vintage/retro video filters using FFmpeg. If anyone is curious, I shared a bunch of my notes on this topic here a little while ago: https://zayne.io/articles/vintage-camera-filters-with-ffmpeg

Disclaimer: I’m still an FFmpeg noob

Site doesn't work well with disables JS because of AMP. please disable this useless crap.

But cool effect without expensive tools :)

To me the film damage/dust/hair overlay seems a bit severe. It’s kind of a default “old stuff looks like this” filter, but I don’t think old films were actually this damaged unless it was stored outside for years or something.

You can reduce its impact by adjusting the opacity while doing the overlay.

mostly it comes from heavily played film from the movie theaters, as the film would get banged up and damaged from repeated plays

Slightly off-topic, but why do old time football videos always seem like they're played at 80% speed?

Do you mean that they look cartoonish-ly fast or slow? It could come down the frame-rate of the camera used for video capture.

Slow. If you look back at footage of say superbowl 1975, everything just moves slower than real life. And they don't make an effort to match the timing of the film to real life time, like when they show clips from 40 years ago or whatever during football games

Does anybody know how to overlay a shorter footage on a longer one? Because the shortest footage just freeze at the last frame in the tutorial. I want the shorter overlaid footage to loop on top of the longer one, Here is what I've tried:

    ffmpeg -i "Old Film Grain.mp4"  -i "main-footage.mp4"  -filter_complex \ 
    "[0]format=rgba,colorchannelmixer=aa=0.25[fg];[1][fg]overlay[out]"\ 
    -map [out] -pix_fmt yuv420p -crf 18 final.mp4

The shortest footage (Old Film Grain.mp4) doesn't loop as an overlay.

I like to get the duration of the base video using ffprobe, then use -stream_loop -1 to loop the overlay video on top of it, while using -t with the duration before the output to ensure it will only be as long as the duration of the base video. Here's an example:

    DURATION="$(ffprobe -i input.mp4 -show_entries format=duration -v quiet -of csv='p=0')"

    ffmpeg -i input.mp4 -stream_loop -1 -i overlay.mp4 -filter_complex "
    [1]colorchannelmixer=aa=0.1[2];[2][0]scale2ref[2][1];[1] 
    [2]overlay" -t $DURATION output.mp4

ffprobe step nor -t needed.

    ffmpeg -i input.mp4 -stream_loop -1 -i overlay.mp4 -filter_complex "
    [1]colorchannelmixer=aa=0.1[2];[2][0]scale2ref[2][1];[1] 
    [2]overlay=shortest=1" -shortest -fflags +shortest -max_interleave_delta 100M output.mp4

Thanks, adding a duration to the CLI args works well.

Try adding "-stream_loop -1" before first -i argument.

thanks, -stream_loop -1 just loops the encoding forever for me, adding a duration with -t fixes it.

You forgot to add a vignette filter :)

Also, the vignette preset for the curves filter is a bit over-the-top, you may want to get a smoother preset (the vignette filter can import Photoshop preset files, so you can look if you can find a free preset somewhere). Alternatively you can find out strategic keypoints within GIMP and translate them to FFmpeg parameter (a bit more hassle I admit), but note that the interpolation between points might not be exactly the same.

Next weekend's project :) Thanks for the tip!

Seems like a cool weekend project to get this whole flow working as a lambda.

Charge people 10$ a month or so to upload and convert videos to this

It would be even cooler to get it working in user browser with local CPU and no upload thanks to ffmpeg on WASM: https://github.com/ffmpegwasm/ffmpeg.wasm

Very very cool.

The only issue is then your depending on the user not doing something stupid during the 5 minutes of so it takes to process.

I personally don't have time for this experiment, but it's definitely interesting

If mega.nz can keep user attention during download and decryption - yes, that happens all client side - then I don't see user doesn't wait for local transcoding process to finish. Show them pre/post processed keyframes as progress indicator.

The last step involves video compression which could take a bit of time depending on the size of the video. If someone can put it on the cloud, I'd love to take a look :)! Cheers!

If only now you could add the film going through the projector shutter sound! (and the reel flapping at the end when done)

I'd add some blur, applied in random spots but large enough not to notice it's shifting.

Reply on news.ycombinator.com