Junction 2022 – make videos great again

16:47, ready for action!


Dall-e is in public beta, so Carlos was excited about using the API

Place to draw&chill

Talking with Luisa from Modash and Dasha from ClickAndGrow.

Super hyped from caffeine from NOCCO

After splitting responsibilities and convincing Carlos to allow me deal with video processing (which he considers really difficult problem) I started my journey: (Saturday, 00:13AM)

As you can see, I’m complete noob

Made an example run from Doesn’t do what I want, but it’s a start.

In about an hour (01:19 I had beautiful video of Carlos over some screen recording I had laying around on the mac:

% time python
python 37.18s user 17.06s system 111% cpu 48.471 total

6 second video. So 8 second per second.

But this was single-threaded work. Can be parallelised, if I would bother. For real-time demo 6seconds is clearly a limit. For submission magic of video editing will help.

Saturday, 12:08. Discussed with Smartly team. Carlos didn’t want to share preliminary demo with Dall-E and other fun stuff.

Went for lunch/breakfast, but door was still closed. Too early and too late at the same time 😉

Found cool repo – Looks promising – black background, should have upload and stream functionality, dockerised. Let’s try to adapt it

The joy of M1 Macbook:

[Warning] The requested image’s platform (linux/amd64) does not match the detected host platform (linux/arm64/v8) and no specific platform was requested

Docker builds super slow

in the end, flask is our everything. Had to fight few issues since I initially used TurboFlask, and had enough dummy UI issues.

At 17:14 I was able to render 6 videos in parallel with overlaying text in different places, and image of user’s choice.

Carlos did a ton of cool AI stuff, now we need to figure out how to connect all of that

We even missed Alpakas around the campus, while trying to make everything work.

It was great to take a 20m power nap in sleepind pod by Reaktor -> their challenge about mental health was so popular – more than 50 teams have chosen to work on it! Talk about competitiveness.

To see how well our software works so far -> enjoy Auto-generated video with AI-generated “energetic jumping dog”

Of course, when partners came to see what we were doing, I got an unexpected error when rendering videos.
Good that I have a “read-only” page where I can view existing videos.

After a bit of cotton candy and prettifying the results page, my energy went down to 20%. So walk at 3 degreee weather was needed.

Time to think about user flows and about what and how to demo.

(and maybe make things look a bit pretty)

2AM Sunday, things finally work together. Awesome AI generational thing by Carlos is connected with video generating pipeline, whole website is navigatable :tada

Video… Just demo of the product will be very cool. Or will it?

3:36. The first 15 seconds of the video make me laugh really hard, with this “talented content creator”.

5:03, feeling strong still.
Carlos fixes all possible and impossible bugs, he’s unstoppable.

Demo efffect – when Smartly partners came, of course we had an error when applying AI generated turtle, and I redid rendering on my laptop and using existing images.

After reviewing 13 projects, my job is done. Lots of projects with cool demos, some with nothing.

Only 2 people came to view us directly, but youtube views were growing (maybe from my insta story)