Eaton Fire: thousands of hours of footage

Apologies for just diving in. I'm not sure where to start with search terms to get a handle on this. If you know of an existing post or resource I could consult with, I'd love to know about it.

Background: On January 7th and 8th, 2025 my house barely made it through the Eaton Fire in Altadena. My Qnap was running through the whole event and the day or two subsequent before the backup power gave out. Fairly soon in the recovery I was able to get it back up and running on generator power. The Qnap keeps about two running weeks of footage from roughly 12 POE cams. I began purchasing 5TB USB drives and archiving off the footage before it was lost. I have had some technical fumbles etc, so there are some gaps in time, but by-and-large, I have been preserving continuous archives of daylight hours for three months. I'm at about 30TB of media so far.

I've been saving the media without any specific idea of how or what I can do with it. My nephew and I both have set up Davinci Resolve 19 on separate fairly high octane computers, and we have figured out how to use Blackmagic Cloud to share timelines each with our own full clones of the media. He's using a new Windows PC that we built last November, and I'm using an M2 Max Mac Studio

We have experimented a little bit with trying to make proxies, and the Blackmagic Proxy Generator App, but the clips are each 12 to 14 hours long generally, and the first crucial day are 24hrs long. We don't want to put a bunch of time and energy into proxy generation without a clear path towards a next step.

I've also been looking at multi-cam editing, but not sure if that would be feasible with the clip sizes we have.

One seemingly good end-objective would be to create timelapse footage from the media. We did one clip and we've looked a little into automating that with python, but again, don't want to start building a workflow without a clear end goal in sight.

Another wrinkle is that the initial clips are in mixed frame rates and formats (generally h264 and h265). At some point I was able to get the frame rates updated on the cams and change them all over to h265 to save disk space, but the most crucial footage is mixed.

Any tips and advice welcome as far as what to do with this media.

What is the footage of, and what is your end goal? Without any info it's hard to know why you have so many camera streams and what those streams are!

He says in the subject that it’s the Eaton fire and it seems like his house burned up. So probably security cameras watching the whole thing 24/7.

We have experimented a little bit with trying to make proxies, and the Blackmagic Proxy Generator App, but the clips are each 12 to 14 hours long generally, and the first crucial day are 24hrs long. We don't want to put a bunch of time and energy into proxy generation without a clear path towards a next step.

I don't think that's a good workflow. I think given that you're apparently working with security cam/CCTV footage, of different framerates, you would be better off actually doing the transcoding manually within Resolve. I also think you should have a clear idea of how the material is eventually going to be used. I would not try to render out 12-hour or 14-hour timelines in one stretch... I would break it up into something more reasonable, like maybe 1-hour or maybe 2-hour segments, which will be easier to load into NLEs (all NLEs, not just Resolve).

BTW, you have my deepest sympathies -- that was a terrible fire and the entire city went into shock when that happened. Just an awful, awful situation. It's going to take years to clean that up and rebuild.

It's not clear to me if you're asking for technical or creative input.

From a technical standpoint, one of the best kept secrets is in Project Settings -> General Options -> Mixed frame rate format. If you set that to None, there's no retiming in Davinci and different videos become much easier to work with.
Assuming that your high octane computers have anything above a nvidia 3060, you can play 4 2160p streams simultaneously in a 2160p timeline.
I'm therefore questioning the need to recode all the footage. (if that isn't possible, then we have an interesting problem and we need mediainfo specifications for the video files and of the computers)

From a creative standpoint, my question is: What do you want the final result to be? Who is your audience?
Because there's a massive difference between "See how my house survived" 2 hour documentary vs 10 minute highlights for social media. (Do note that I'm not saying what way to choose!)

I thank you all for your suggestions. This point from Mads Johansen is a great question:

It's not clear to me if you're asking for technical or creative input.

It's more the creative side first, with an eye on the technical implications.

I have been archiving the daytime footage on instinct because it seems like a moment in history. The only way I've thought of to make the vast volume of footage any use is to generate it into time lapses. Depending on how automatable this can be, I'd consider timelapsing all of the footage, but if it's too labor intensive, I could definitely be more selective about which angles to prioritize. This is a frame grab of the views I have and a couple examples of the views I captured the night of the fire.

Untitled 1.jpeg (351.21 KiB) Viewed 1265 times

Untitled 2.jpeg (114.83 KiB) Viewed 1265 times

Untitled 3.jpeg (134.85 KiB) Viewed 1265 times

Obviously the ones with active fire are a singular consideration, and since it's a 24hr period, not a massive technical challenge. The subsequent footage chronicles the aftermath and recovery. I have an instinct that there's something worth creating with that. What are your thoughts about the utility and value of the aftermath stuff?

I have a few scattered thoughts, without seeing the footage I'm less directed than I'd have liked to be, sorry.

Initially I didn't like that there were timestamps in the videos, but after thinking about it for a while it makes the passage of time simpler to show.

Without knowing the details, my creative process would be to show some highlights: 1) First sighting of the fire, 1.5) the force of inevitability, 2) people leaving, 3) inferno, 4) inferno leaving, 5) people coming back.
If there's police, ambulances, fire trucks, lumberjacks cutting trees, helicopters, show them.

Timelapses can show the speed of the fire, to show how little time people had to react.

As for the aftermath: It's basically cleaning the area, right? Is that something people want to see?
Without seeing the footage, it's difficult for me to envision it. I think that's why I'm not convinced it's worth the viewers time.
(It's partly a challenge btw. Show me I'm wrong!)

My frame of reference is this video:

Hey dualmon! I'm a doc filmmaker from Altadena... my house didn't make it unfortunately. I am working on a feature about the Eaton Fire and the recovery and am really interested in your footage! Let me know if you are interested in talking.

Thanks for the input. FWIW, we're now experimenting with pre-processing the normal-speed clips with ffmpeg to make timelapse versions. I have a diary of events, so will be able to use the diary and other footage to identify days when something potentially interesting was going on, and then go through the raw CCTV footage to see if anything worthwhile is in there. If anything public comes out of this project, I'll share it here.

Hey dualmon! I'm a doc filmmaker from Altadena... my house didn't make it unfortunately. I am working on a feature about the Eaton Fire and the recovery and am really interested in your footage! Let me know if you are interested in talking.

@sonic0_0tracer I'm sorry for your loss. I would love to talk. Does this board support DMs? I'm new here. My posts are moderated.

Regarding the ffmpeg technique, I have found a good set of parameters to take it my clips from 14 hr to 5 min. It takes about an hour fifteen to process one file and results in a file size down from 32GB to 198MB at 5400k bitrate and 30 fps.

Code: Select all

ffmpeg -i input.mp4 -filter:v "setpts=0.00595*PTS,fps=30" -an -b:v 5400k output.mp4


ffmpeg settings.jpg (191.78 KiB) Viewed 1046 times

1/168 is an unusual timescale, how did you come up with that number?

Personally I don't like -b:v because it set an average bitrate which may or may not be enough for the footage.

I much prefer -crf 18, and -preset faster as in this example:

Code: Select all

ffmpeg -i input.mp4 -filter:v "setpts=0.00595*PTS,fps=30" -crf 18 -preset faster -an output.mp4

With crf you don't know the final size, but you know the final quality, with crf 18 being visually lossless. -preset faster gives the encoder less time per frame to do the compression (and as such would be a death sentence for quality with -b:v), but since we have demanded a certain quality with -crf, the encoding won't suffer (yes, the final file will be larger compared to -preset medium, which is specified when nothing else is specified. -preset slower and -preset veryslow are slower but less file size)

Have done similar work in the past for the client that wanted a timelapse version of his 24h clips from security cameras to something short and more usable for him.

Have used a SnapMotion app that automated all process. You can set various parameters for extracting stills from videos (png, tiff, jpg..) interval of extracting etc..
You can have all images or the app can create a timelapse video for you with you setting all paramters of interval of capture etc..
So if that is something that you are interested in check SnapMotion on appstore.