VP Land
Posts
Strada: Building the Road to an AI Powered Media Workflow

Strada: Building the Road to an AI Powered Media Workflow

Joey Daoud
January 26, 2024

In this episode of VP Land, I have a wide-ranging conversation with Michael Cioni, CEO of Strada.

We talk about the future of media workflows, how AI will help us work faster, and why he decided to build his company in public.

YouTube | Apple Podcast | Spotify | Overcast

Episode Takeaways

Strada's vision is to use AI as a utility to speed up and enhance media workflows. There are plans for a custom AI marketplace with a variety of models to solve or complete specific tasks. But that comes later.

The 4 Ts of Strada

At launch with their private beta on Feb 1, Strada is first tackling what they call the Four Ts, to create a foundation for working with media.

Transfer

This is about getting the media off the media cards and onto storage. Strada is not a storage provider or offering any cloud storage - its goal is to connect to existing storage, starting with Dropbox, Google Drive, Frame.io, and a few more.
A big announcement Michael revealed in our interview was not only will Strada work with cloud storage, but it will also work with connected hard drives, from USB drives to NAS.
Anyone with permission and access to the storage via Strada will be able to download the media directly from the hard drive without having to go through a cloud intermediary.

Transcode

Transcode is a combination of converting or debayering media to other formats that might be needed for editing, and for linking similar assets together.
They call this a Strada Stack - a virtual file that uses timecode and other metadata to sync multiple assets together, from camera and sound to additional angles, behind-the-scenes photos, alternate versions, LUTs, and more.

Transcribe

Transcription is usually thought of for long-form interviews, but Strada wants to normalize transcribing all footage, since it's the easiest way to find moments in takes and search through footage.
It will also tag speakers - and remember their voice. Tag someone once, they'll automatically get tagged in other clips they're in.

Tag / Analyze

And the final T for Strada is tagging and analyze. This is where AI really comes into play to analyze shots for objects and setting, so footage can just be searched and doesn't need to have a rigid organization system.
Michael put a big emphasis on being able to be stay messy and creative and not have to rigidly tag all your footage to find it. Their goal is AI will do all the searching for you.

Cloud Processing

Strada is built to take advantage of the cloud for processing and rendering...if you need to. In this demo from one of Strada's YouTube videos, they were able to use a cloud cluster to render a batch of video clips in parallel instead of sequentially.

If you have the time, it seemed like from our conversation that you can run the processing on your local machine for transcribing, tagging, and transcoding. But if you're on a large shoot and speed is important, you cnan buy cloud compute time, even reserving virtual machines for your project's exclusive use.

From Strada to NLE

Strada will initially let you pull highlights and selects from clips, build out a paper edit or collection, and send via XML to your editing program.

There was also mention of allowing basic trimming of clips, but that won’t be released in the private beta.

Changing SaaS Pricing

Strada understands that we all have subscription fatigue. They're rethinking the pricing model, especially for creatives where projects and workload are not consistent year-round.

One way they’re changing the model is by not charging for seats. According to Michael, you can bring as many users into your Strada workspace as you need.

The pricing seems to be a base monthly fee for an account (not yet disclosed) and paying for cloud compute time when needed.

AI Marketplace

Yes, everything seems to have AI today. But Strada has an interesting approach to their vision of the future.

A marketplace with lots of small, custom-built AI models that do a specific task. This is more like HuggingFace, the AI model community platform, less like another OpenAI/ChatGPT wrapper.

Smaller, more specific models are an AI trend that we also discussed with Jose Puga of Imaginario.

Michael gives a few possible scenarios: you film on a noisy street and need to remove car noises. Instead of running a generic noise removal tool, there might be a specific model trained on just removing street noises.

Or you need to hide a logo on a t-shirt in a bunch of shots. You could use a model to erase or replace the logo after scanning through all your footage and transcoding new clips with the logo altered.

This would have an app store model where developers can post their own AI models, set a price, and Strada takes a cut.