@mux/ai

I would like wake

up and I would

hear the, the like

support chat sound

notification from

our website and

go, Oh, great.

I got to go fix

like a video

problem again.

And it would

just take up

hours and minutes

of my morning.

Every morning Mux

enables us to not

worry about all

that junk anymore.

And I can

actually focus on.

Our strengths,

right?

Like I can build

cool things again.

Meet Chris Sev.

He's the founder

of Video Tap, a

powerful app that

makes repurposing

video content

easy and fast.

Chris recently

grabbed our

attention when he

tweeted about how

a switch to Mux

not only improved

video taps,

video processing

speed by 80%.

It also made it

much more reliable.

In this chat, we'll

talk about what his

pain points were,

how switching to mux

helped solve them.

He'll share his

insights on building

apps in the crowded

AI space and more.

Let's get into it.

We worked out our

old and buggy way

of handling videos.

Our new video

processing pipeline

is 80 percent faster

and more reliable.

Okay.

This was a tweet

that went out.

Can you expand

on that?

Like how, how

did you do that?

Or what was the,

what was the

friction that you

were running into?

And then how was

that resolved

on your side?

Yeah.

So.

I'll tell you

the stack and

then I'll tell

you the friction.

So the stack was

we would upload

videos and I did

this because I just

wanted to stay as

cheap as possible.

Video can get

expensive pretty

quickly, right?

You all know.

Uh, so we upload

to CloudFlare R2.

So that's kind of

our main storage

point there.

If we had a

user upload a

video, we would

also pass it to

CloudFlare Stream.

Which would

encode it for us.

And then we would

kind of re upload

it back to R2.

That whole process

sucked to generate

the clipped

sections of the

video, because we

generate clips.

So basically if

you're trying

to clip like 30

seconds to one

minute, right?

Like a 30 second

clip, we would

pass it over to

AWS media convert.

And that would

grab the video from

CloudFlare R2 and

then like clip it at

the right section.

And then it would

store that on S3.

And then.

I didn't want

like two different

storage locations.

So then I would

just move that

over to R2.

And then the last

part of this is to

render the clip.

We're using a

react package

called ReMotion to

do all the video

editing and, and

clip rendering.

So ReMotion

does the render

on AWS Lambda.

There was like

a hundred points

where this thing

could break and to

debug it was just

a mess every time.

And it would

always break.

Like it got to

the point where I

just like almost

quit the project.

Cause I like

didn't, I couldn't

figure this

whole thing out.

Right.

So.

When you're thinking

about the, the

changes that you

made to the pipeline

and it being eight,

when you, I think

it says 80 percent

faster, was that

like considering

the development

time or the, like

the frustration or,

or like the time

to you actually

shipping a feature?

Or is that

like, actually.

Clip being made

and generated is

now like measurably

faster as a result.

Yeah, no, it's,

it's after a video

is uploaded, how

long does it take

to become ready

to be clippable?

And then how

long does it take

to, to actually

generate each clip?

And both of those,

I said 80 percent

to be a little, I

don't know, like

conservative, but

it really is like,

90 percent faster.

My stack is Laravel

and to do the

whole change up

was not that hard.

Cause as soon as the

video was added, I

just have a chain of

actions that happen

and mux did all that

stuff and then it

just sent back a

webhook, which was

really nice because

none of the other

things had webhooks

so I was just like

polling APIs over

and over and over.

And the code is

just much simpler.

Now it's very easy

to understand the

overall flow of how

our project works.

Chris, maybe you

can give us a

quick tour of how

VideoTap works

under the hood.

Sure.

So, I am on the Mux

YouTube channel.

Basically, pull

the latest video,

optimize your

website's video

for social media.

Just grab the

link here.

If I go add videos,

and I'm going to

add from YouTube.

We'll paste

in that link.

Okay.

So, everything

is good here.

You click start

processing.

It jumps over here.

And then, we do a

couple steps here.

As far as processing

your video, this

would take like

estimated time.

Here's one minute

that actually

is pretty close

to accurate.

Um, but before it

was like, it's a

four minute video,

as you can see here,

it would probably

take us about

like 15 minutes

to process and

people just, you

know, want their

content faster.

Right.

We have two split

things happening

at the same time.

We have the

video processing

through Mux.

We have.

Transcript

generation done

through assembly

AI assembly.

AI is actually

introduced a new

model, made it

super, super fast.

Like you can

transcribe a 10

minute video in

three seconds,

and now it's able

to go generate

title chapters,

clips, blog posts,

and all that.

So between assembly

and the mux update,

like these two

things alone.

Used to take 10

minutes on a four

minute video and

we're on this call.

Like it's

already done.

It'll generate

the clips and then

it'll kick us over

to the video page.

But while that's

working, let

me jump into my

MUX dashboard.

Yeah.

So before we would

on this page, this

processing video

would be like a

bunch of steps,

it would basically

upload the video

to CloudFlare

R2 as storage.

Use that to send

that video over to

CloudFlare stream

to encode it.

And then we would

re upload it back

to R2, uh, and then

delete it off of

stream to like save

the money there.

There's a lot of

places where like

I did stuff just

to save money.

Looking back, it's

like such a pain,

but like it, I

get why I did it,

but it's gone.

Now I don't have

to do that stuff.

So we're able

to kind of get a

title here, get

your description,

get your chapters.

You know, AI hit or

miss sometimes, but

I think we've done

a lot of work as far

as like prompting

design and just

getting the correct

information into the

prompt that I think

these chapters are

pretty accurate.

Uh, tags are

pretty good too.

Something that I

want to do here

is pull in from

the YouTube API

as well and see

like current tags.

I want to do

generate titles and

then compare across

like different

competitors

type stuff.

A lot of that's

down the road,

but I think that's

going to be really

helpful as far as

like just being

more competitive

on YouTube SEO.

Summary, which a

lot of people like,

um, it can also lead

into like podcast

show notes, which

is a feature that

we're working on,

but it's pretty

close to this and

a meta description

for your blog

post type stuff.

So here's the

blog post.

And sometimes

it gets a little

wiry with these

code blocks, but

one of our cool

features here is

like, you can add

a screenshot, uh,

drop it in directly

from a specific

point of the video.

Right.

And that also was

done with mux,

which made that

like a two second

process before we,

what would we do?

We would get

the video from

Cloudflare R2.

We would use FFmpeg

to screenshot

that section that

would create an

image locally

on our server.

And then we

would upload that

to R2 as well.

I'm not even

trying to like

hype you guys up.

It's just amazing.

So this one turned

out pretty good.

Right.

And a lot of the

stuff here, uh, the

next step for our

blog post feature is

giving you options

to like tweak SEO

keywords, to rank

across different,

uh, things, kind

of like the tools

you'd find at AH

refs or, uh, SEM

rush type stuff.

So that's another

thing down

the pipeline.

Um, A lot of things

that I want to

do that I haven't

been able to do.

Cause I'm out

busy, like fixing

video bugs, but you

know, monks is able

to like, give me

more time because

all that's all

that's fixed now.

And I guess our

big feature, nice,

good screenshot,

our big feature at

the end is clips.

We give you a title,

kind of what's good

about this clip.

And then the

specific texts

that you'll show

in the clip.

And so I think this

is kind of like one

of the wow moments

that people see.

And we give you

a score on it.

I think there's a

lot of people doing

clips nowadays, as

far as like the AI

world, because I

think there's like a

lot of value in it.

Descript's doing it.

Like a lot of

these big players

are doing it

where I want to

differentiate

here is.

Better styles.

I think we can

do some really

cool stuff.

And then, uh,

overall workflow,

like it would

be amazing.

Like one of our

goals here is to

have a publish

button and then you

click publish and

you see like the

four clips that you

have, and you can

just drag and drop

to like Monday at

10, Tuesday at like

11, uh, Thursday

at whatever.

And then just

publish all four

of these to like

TikTok, Twitter,

uh, LinkedIn and

YouTube shorts,

Instagram reels,

wherever, but do

all that from, from

So as far as style

changes go, we

have a couple of

different like

styling options.

Um, yeah, you

can edit your

clip texts.

We're trying to

write social media

posts for these

types of things.

I think this thing

can be like editable

in the future

and tweakable.

And another thing

we're working on,

uh, is, uh, like

match your tone.

So if you connect

your Twitter account

and your social

media accounts, we

should be able to

like look at your

previous posts.

And that's something

that, uh, we now

have more time

to work on again.

One of the cool

things right here

is this is the main

video at four 21 on

the duration here.

Right.

And if you scroll

in, we added the

video at like 1250

dot 24 and it came,

it was ready at 52.

Right.

So like 25 seconds,

the first time I

saw that was like

the big point

where I said, Oh,

this actually can

change a lot of our

user experience.

Each of the specific

clips, all of

these correspond

to one of these on

this page, right?

Each of those clips,

like they were

pretty much ready

in like ready

to stream in two

seconds and then

ready to download in

20 seconds, right?

I get the appeal

of building out

your own workflow

for video.

Like I had a server

that was just

doing downloads and

FFmpeg type stuff.

There's just so

much overhead in

a lot of different

areas that.

If you're trying

to build something

amazing and

something really

cool, if that

stuff is not your

differentiating

factor, don't do it.

Folks like you

in a situation

where you're

trying to build

something great,

but starting from

kind of nothing,

like starting from

scratch, we need

to be there along

with you, uh, along

in that journey.

That we have a, some

infrastructure in

place that solves

all those pain

points, but doesn't

break the entire

budget because

you cannot invest

entirely into any

software out there

without having

your business model

proven, right?

So like, yeah, we,

we have to have

options available

for folks that

are in the exact

position that, that

you're describing.

Yeah.

And you did it.

Like I looked at

it and immediately

thought no way

that's real.

And I think that's

a good thing

when you look at.

Which, which part,

can you, can you

go into a little

more detail there?

Like what plan

did you end up

choosing that's,

that's fitting this

particular use case?

I'm on that, I think

it's what, it starts

at 10 a month and

then it's like usage

base after that.

Got it.

Is that right?

Yeah.

Yeah.

Nailed it.

Yeah.

So for me, a lot

of my use cases,

clipping and, Just

video processing

and coding at this

point and not really

streaming right

most most of our

people that generate

clips They're gonna

download them and

upload them to

social wherever they

go So, you know not

having any anything

where we're paying

for things that

we'd like like,

oh darn I wish I

didn't pay for that.

It's all like

perfectly placed

what you consider

for this This

product isn't just

what a lot of other

people consider

for an AI feature,

which is like,

throw it to an LLM.

Uh, that's really

like, we kind of

take off the mask

and what's happening

under it all is

like, all the hard

work is happening

on an LLM and that's

all that we do.

Uh, but that's

not what you,

you all do that.

You, you all do a

lot more than that.

And where, where

do you foresee

standing out?

Um, not only

with your AI.

Integration and

implementation, but

in the way that you

talk about it in

public, uh, with

all of the other AI

noise that's, that's

out there right now.

Yeah, yeah,

you know, AI is

everywhere now.

Like everybody's

either creating

something new with

AI or integrating AI

into something old.

Right?

So at this point,

I like our, our URL

used to be videotap.

ai.

And I deliberately

said, no, let's

go get videotap.

com because in

a year from now,

everything's

going to be AI.

So what does

it matter?

Right?

I don't even want

to put that really

in our messaging

anymore, because

it's kind of assumed

at this point.

Like I don't

care if it's AI.

I just care that

it made like

good content

at this point.

I saw a great tweet

from Paul Graham and

he said that for AI

tools, your biggest

differentiation

isn't how you use an

LLM, your moat, like

your differentiator

is always going to

be integrations.

So that's where

we're leaning.

Now that we're,

we've, we're done

building the core

and especially

with mux, I'm,

this is the first

time I feel the

core is stable of

our foundation of

what we've built.

So now we can

build on those

integrations.

And I think

that's where.

A lot of people

should be looking at

for their AI tools.

If I have a 10

different Zapier

zaps that are

integrated across

like my Google

sheets, my like

calendar, all

these things.

I'm not

leaving Zapier.

It's going to take

me a long time.

Right.

That's the

flow here.

So a lot of

our clients are

marketing teams that

want to use this.

And they're,

they're like

really want that

publishing feature.

If we can streamline

that whole flow

for them, like they

don't care if it's

AI generator or not.

They care that it

like saved them

that workflow.

It seems like Chris,

you, you spent a

lot of time and

investment on in

the community in

some way, whether

that's just like

building in public

or sharing what

you know, through

tutorials or what

you just learned

or your experience,

what, what value

do you see in does

it, does it help

or I won't even put

words in your mouth?

What do you see in

the contributions

out in the community

on, on the internet?

Video tap as, as

a SAS is it grew

pretty quickly.

And we're at a

point now where I

think the skillset

that I had as far

as just coding and

just being like a

little, you know,

in the cave coding

and just getting

things out there

and getting things

done, it grew us to

where we are today.

But I think that the

level where we're

at, that just won't

do the job anymore.

I have to like, get

out there, talk to

people, be a little

bit more proactive.

And I think every

company comes

through this,

like, You have to

talk about what

you built phase.

If, if you're like

proud of what you

built, then why

wouldn't you go

out and like tell

people about it?

Right.

And I don't

necessarily

excel at that.

And I always think

of Twitter as like

a distraction.

So there's a lot of

times where I think

that I'm wasting

time out there, but.

Like seeing what

other people are

doing, being able

to talk to you,

being able to, you

know, chat with

other people that

are building the

same thing, have

the same problems.

I think if anything,

it's just good

to have the water

cooler and talk to

other people that

have your problems.

I haven't really

tracked metrics on

it and like, and

I don't know if

that's a hundred

percent the way

to look at it.

Like, Oh, you know,

50 people came from

X this week, none

of them converted.

It's, it was

a failure.

It's, it's top

of funnel, right?

Like you can't

really track

how those people

convert, but brand

sentiment, how

they feel about

you, how I feel

about mux, right?

Like it goes a

long way, uh, that

dev love ability.

Uh, louder for the

people in the back.

Um, anything else

on your mind?

Always shout

out my wife.

She's fantastic.

She, uh, just

started a new blog

called richest you.

She does a

lot of like

personal finance.

So if you want to

check that out,

she's also in DevRel

actually, and is

a video tap user,

the most critical

video tap user.

Like she's over

here at like her

desk and she'll

just say, Hey, Why

isn't this working?

And I'll have to

like go over and

she's like clicking

a button that

isn't should work.

So that's, that's

a great way to

get feedback.

yeah, I don't know.

This was fun.

Thanks for

having me.

I enjoyed

implementing Mux,

using Mux, and

I'm excited for

being able to

build stuff again.

Let me ask you

one last thing.

Where can people

learn more about

trying out VideoTap

or just about you

on the internet?

Yeah, so VideoTap.

com is the spot.

We had to go bid

for that domain to

get the dot com,

but we got it.

Uh, And on Twitter,

xchris__sev,

chrissev was taken.

But, those are the

two big spots that

you can find me in.

That's really

where I, like, make

the most noise.

Awesome.

Thanks Chris.

Thanks.