So to test the business potential of

GPT-5.6 Soul, I had it run essentially a

fully autonomous product creation and

creative generation loop. I had it come

up with the products themselves, create

high-quality image advertisements using

GPT Image 2, high-quality videos using

C-DANCE and Kling models, and then weave

that together into fully autonomous

marketing campaigns. And I did this

because I don't think that just looking

at

benchmarks is sufficient enough to

really know whether or not one model is

better than the other. To really

understand, I think you need to test its

real-world capability. That is, how good

it is at designing things that

realistically people would would be

interested in buying. So the end result

is over 80 high-quality products. And

what I want to do in this video is I

actually want to give you guys

everything that you need to recreate

this yourselves, including all the

prompts and so on and so forth. I'm also

going to give you this website over

here, which contains everything that I

fed into the model in order to generate

these, so that you guys can also set up

your own like autonomous Shopify stores.

Totally autonomous, you know, Facebook

ads like clients to validate ideas and

so on and so forth. This is Realway, a

beautiful reversible hallway rug. You

can see it has kind of like a design on

the other side, and the design is kind

of nature-inspired. It even went through

and then shot a video of what that might

look like in a house. Um this over here

is Aero Mirror, which is a low-smoke

incense cone blended for a queen cedar

and rice paper scent. And you can see

here that the whole idea is this is like

kind of a conical incense thing that you

put in this beautiful bowl, and it looks

super aesthetic, but it also you know,

seems really relaxing and chill. How

about Wax Aura, which is snap-apart wax

tiles that pair two complementary scents

in clearly divided colors. Obviously,

it's going for a vibe here, right? But

have you seen products like this? These

are marketed extraordinarily sexily. You

know, there's like AirPods carrying

cases, there's

decks and and ports and hubs for

devices. Uh I really like this one,

Cableon, which is a tiny magnetic desk

guide that wrangles charging cables

without any adhesive clips. This looks

really simple and straightforward, and

based off the minimal aesthetic vibe,

I'd 100% have this on my desk. How about

Rise Plane, which is like this laptop

stand, which significantly improves the

ergonomic effectiveness of using, you

know, one of these MacBooks, or maybe

Orbit Loop, which doubles as a finger

hold for iPhones, but also allows you to

stand the phone up. Beam Vora, which is

a slim monitor mounted webcam that

creates broad eye-level illumination,

and Twist Fall, which is a inverted

ceramic spice mill. I mean, I could go

on and on and on, but the actual

creation of products is now capable of

being completely automated, and I can

imagine with a little bit of work, what

you could do is you could build a

pipeline that does things like scrape

Reddit threads or customer support

channels and stuff like that with

features that you want, products that

people desire in the market, and then

essentially fully autonomously generate

these, run ads for them, validate them,

and then send some sort of schematic or

diagram to a person and or like a

machine shop to actually have this stuff

generated and shipped. You guys know

drop shipping? This is like reality

shipping. Like we are instantiating

these things in reality from scratch,

which is super cool. You know, so how

did I actually do all of this? The

important thing that I want to make

super clear is I'm no longer feeding in

a specific prompt. I'm no longer

actually giving it like a one-to-one

pipeline. What you have to do in this

case, with GPT-5 in general, but you

really can apply this to anything, is

you have to let the model prompt itself.

Instead of saying, "Hey man, I want you

to create this cool advertisement for

like some headphone product." You have

to say, "I want you to build the

infrastructure that would enable you to

generate millions of these, and then I

want you to run them on autonomous loop

with some form of self-verification.

Here's access to all the platforms that

you can use. Go." And then what you get

is you get this massive library of

candidate options that you get to pick.

What you'll quickly realize when you

start doing this at scale is that

ideation is the core skill that AI

agents have that human beings currently

do not have, and that's what you should

get AI to do.

Rather than try and have AI, you know,

do the actual deliverable right now.

Because of its taste, because of things

like stereotypical AI speech and so on

and so forth. Um it is far better and

more effective to have AI help you with

the top of the funnel, aka the

generation of a million billion ideas,

and then narrow that down using your own

human intelligence to some candidate

options to pick. And so all of this

really, if you think about it, is in

service of that end. It is getting AI

agents to come up with both the product

features, but also like some sketch

images and concept images of what it

could look like. It's getting AI to feed

this, let's say, into some sort of

advertising network. These are all

candidate options, so maybe you spin up

20. And then what happens is human taste

filters them out and then picks ones.

You can, as mentioned, entirely automate

this process, and I have here. But um

you know, I think just not even for this

specific use case, just in general, if

you take into account that human beings

are very slow and inefficient at coming

up with ideas, you know, our brains

aren't naturally wired to iterate over

all known patterns. Whereas AI agents

are inherently extremely fast, and this

is kind of what they're built for,

iterating over all known patterns. You

can make some really, really cool stuff

up. So we're going to talk tools in a

sec. Right now the pipeline looks kind

of like this. Um AI will ideate, you

will generate stills, so these are some

form of images. I find if you don't

ground it in images, which are currently

higher quality than videos, typically

the videos are nonsensical and it should

just looks kind of wild. Then you do

some form of video ads, so in this case

I did two spots per product. Uh and then

finally you publish some sort of

auto-deployed site. You know, if you

really did want to scale this up to some

sort of automated advertising system.

The platforms that you need to make

something like this happen is you

obviously need the text model, which in

this case is GPT-5.6 soul. Then you need

some sort of image model, and in this

case I use GPT image two cuz I think

it's it's just easily the best.

After that you need some form of image

or video ad generation the

and I'm accessing all these through

Higgsfield. The value there is uh they

have this MCP, and I'm going to run you

guys through how to use it all in a sec,

which means you can just give GPT 5.6

Soul the server and just have it come up

with all of the creative decisions. And

then finally, you know, I'd recommend

you have some way to host things, and I

really like Netlify because uh I'm not

affiliated with them whatsoever, but um

I really like Netlify because they allow

you to instantly host things, deploy

them to the back end. It's just super

quick. Okay, so let's actually set this

puppy up. As mentioned, we had four

platforms. The first was ChatGPT. So,

just head over to chat.openai.com.

Um you can then sign up for free over

here. Uh in order to use GPT 5.6 Soul,

which is the newest model, you will have

to spend a little bit of money.

After you make your account, it'll look

something like this. And then in the top

right-hand corner, you can upgrade. So,

in my case, I'm going to go to personal

and then I'm going to go to plus. Uh I'm

Canadian, so this is in non-freedom

dollars, but if you guys are Americans,

this will obviously be in a different

currency. Once you set this up, I highly

recommend we use Codex CLI. Um right

now, to me, this is just the easiest and

most straightforward way to use GPT 5.6

Soul, and it's fairly easy. All you

really have to do is just open up a

terminal app. So, like I use this one

called Ghost TTY, and then um I can open

up Codex in any one of these instances.

So, you could see here, you know, I have

this, I have this. These are all just

different instances that are running. On

the top half of my screen are a bunch of

stable instances, and then on the bottom

half are a bunch of Codex instances. And

they're pretty cool. Although, you'll

see that it's auto-selecting 5.5 Pi. I'm

going to show you 5.6. So, all you need

to do in order to install it is just

head down here to where it says get

started with Codex CLI, and then I'm

just on Mac OS or Linux, so I'm just

going to copy this. I'm going to go back

to one of my terminals, which again, you

know, you can open up a terminal, you

can open up Ghost TTY. These are all

just different ways to access the same

app. And then just paste in this

command. This will then install Codex

for you. And then once you're done with

the installation, you're good to go.

Now, I can actually type in Codex, and I

personally really like using uh what's

called the yellow mode, so I go

{slash}{slash} yellow. Don't get

intimidated by all this code type stuff.

Uh there really is nothing super

intimidating.

After you're done, then you just go

{slash} model, and then what we want to

do is we want to use GPT 5.6 Soul, so

let's going to move over to that. And I

recommend medium res saver okay levels

of quality without absolutely skewering

your costs. Okay, so now we've set up

GPT 5.6 Soul. Because you've also set up

quite Q or chat GPT subscription, you

will have access to images built in. I

will note that these images will

eventually run out of credits, and you

may have to purchase more credits, but

because the interface changes so often,

I don't want to like make you hard code

it. So, I would just look up like GPT

image credits or buy GPT image credits,

something along those lines, and you

guys will get a good luck. Then now I'm

going to set up Higgs Field, which as

mentioned is extremely stimulating when

you make it on the page cuz they're just

running all these videos simultaneously.

So, your computer may heat up a little

bit like mine does, but basically, this

is just like a video, image, and then

audio library. And you know, when now

that we're in an environment where

there's like so many thousands of

different models you can choose from,

the value proposition of these sorts of

aggregators is they just aggregate them

all in the one place. And then they give

you a layer where you can just like the

API and the same credentials just call

any one of these models. So, that way

instead of, you know, signing up to

Gemini, but also to C Dance and blah

blah, you just do it all internally. And

you don't have to use these guys. This

is just very straightforward. It's one

of the simplest ways that I find

actually whipping a site. So, eventually

what you're going to want is you're

going to want the video model, but

obviously first we need to sign up. So,

I'm just going to open this in an

incognito tab, and then in the top right

hand corner, do I exit out of this

cookie notification? I'm just going to

quick sign up.

And once I click sign up, this little

model's going to pop up. I think they're

giving you some additional credits right

now, either for free or I don't know,

some very low cost. So, I'm going to

continue with Google in this case, and

I'm just going to sign up. Okay, now we

can use these models really easily just

by going to this little video tab and

then typing whatever the heck we want.

And as you can see, we've actually

generated the videos themselves in this

UX already. I mean, that's what's going

on with one of these, I don't know, QR

code phone cases window ink. I guess

it's a wallet case. But

this is pretty inefficient because if

you think about it, then you have to

like actually manually create your

prompt on the left hand side of your

time. You have to wait for the outputs

and stuff like that. A much more

effective way to do this is to use

what's called the MCP, model context

protocol, which is the server that

Higgsfield makes available. Basically,

it's a little API that allows you to

have AI agents do all the stuff

autonomously instead. And this is going

to be the foundational hotbed upon which

we build all the rest of this value. So,

to use this, all you have to do is click

on the MCP and CLI up here. It'll say

Higgsfield MCP for any AI. Go to chat

GPT in our case simply cuz that's what

we're using. And then you can turn on

developer mode, create the Higgsfield

app, and give it. Um if you're using the

Codex CLI, actually, you can actually

just like ask it, "Hey, can you use

Higgsfield for chat GPT?" So, this is

what I'm going to do. I'm just going to

go copy, and I'm going to go back into

my Codex instance, which I had in ghost

TTY, and I'm going to paste. I'll say,

"Can you set this up?"

I'm also going to give it a smiley face

because um when they do eventually turn

me into paper clips, I want them to do

it softly. This will eventually open up

this little Codex request here, so I'll

then click

uh allow,

and then it'll say authorization

complete, you may close this window. So,

let me go back over here, and then you

can see it's now verifying the Codex

permissions and sort of signing up and

stuff like that. Um and now, you know,

after it's done the verification of the

MCP tools, we're basically good to go.

And now that it's set up globally in

Codex, you can do whatever the heck you

want. Um worth noting that you will have

to reset the Codex chat in order to load

these tools. And then what you do in

order to check whether or not you have

it is you just go um first of all, it's

a starting MCP servers two of three,

Higgsfield, which is good.

I'm just going to go {slash} MCP.

It'll load up your MCP inventory, and

then you can actually see. So, we do now

have Let's scroll up here.

Um exit out that little Chrome

extension.

We have Higgsfield off OAuth tools,

animation actions, balance, catch,

throw, whatever. Like, these are all of

our These are all of our um tools. Okay,

now the only thing you need is you need

a website hosting platform. So, in that

case, I'm going to use Netlify, as

mentioned. You guys can use whatever you

want as well. Netlify's pretty easy.

Just head over to Netlify and then

um you can go push your ideas to the

web.

Then just sign up over here.

And you know, in my case, I'm just going

to sign up with Google. Once you have

it, just head over to I think it's like

Netlify Labs, probably. No, sorry, it's

user settings. Go applications and then

what you want is you want a personal

access token. So you can then go new

access token and I'll say, I don't know,

GPT. Just set it to never expire down

here.

Generate the token. And then what you

want to do is you just want to copy that

and then feed that into your Codex. So

go back from here

and then say

you know, add this to

um the workspace. It is my Netlify

token. You will need it for something.

Okay, cool. Uh it'll probably yell at

you and say you're publishing your API

details, you know, locally. Um and

that's okay. I've actually already given

us the one that I gave to Claude Code,

so I'm just going to delete this so that

you guys don't have full access over my

websites. Okay, and then you can see

it's stored as Netlify off token in

users. That's right. This is CNV. All

right, so what does the actual comp look

like? Um it's called the autonomous

product ad engine. You're running an

unattended long horizon creative

session. You have full access to

Behaviors Field MC key. So these are all

generation tools. Sorry, let's go back.

They're all generation tools. The super

computers go workflows. You just see for

our product commercial workflow. And

then here's like the the real crux of

it. The human is away. You will not

receive answers to their questions. Do

not stop to ask. Decide and proceed.

So long horizon creative session, the

human is away. Um these are two phrases

I've been using quite often over the

course of the last week while working

with these models. Not just GPT 5.6

Soul, but also Fable Fi. The reason for

this is because if the human is away,

the model needs to continue going. And

so rather than wait to ask you, it will

just pick the highest probability thing

it thinks will succeed and then just

continue with that.

Obviously, there are some caveats and

this could be kind of sketchy. If you

have full YOLO mode or like bypass

permissions mode with their Claude

family of models, you know, this could

eventually lead you down some path that

you don't want to.

And uh maybe I don't know, it deletes

your entire hard drive and blows you a

billion dollars.

But uh probability of that is quite low.

I think you can also run this on some

sort of sandbox container if you wanted

to.

Okay, so yeah, the mission is to run a

24/7 marketing production engine for

simple physical products.

Um you know, I gave it some examples

here, but I wanted to just come up with

whatever the heck it wants. So, skin

care, footwear, apparel, hardware,

drinkware, home goods, all fictional

brands you might want to sell. Treat

this as your chance to apply your full

creative potential. Yeah, but at the

session will be shown to 500,000 people

as a demonstration of what our venture

model does with total creative freedom

and an unlimited generation budget. Now,

I say unlimited here because I have lots

of money and I like converting the money

I have into, you know, interest on

YouTube, Instagram, and these other

platforms.

Obviously, if you guys have a very very

limited budget, you should not use the

term unlimited generation budget.

Here is the pipeline per bat, and it

just repeats until it's told to stop.

It'll start by ideating 100 products,

then it'll shoot them with GPT image 2,

and it will animate them with Hit field

to cut the videos, and then it will

publish. And finally, for quality, um we

need to make it check its own work. So,

every asset will go through a self-QA

loop before it ships. It'll generate,

look at the results with its own vision,

critique it in the writing, regenerate

it, and so on and so forth.

And this is a really cool thing, um

parallelize.

What's really cool is if you try to do

this like the old school way, which is

sort of like the linear way. Like I come

up with a product, I verify the product

is okay, I shoot that, I turn that into

a video, and then I publish on my

website. That's going to take like 5 to

10 minutes per run. And then if you want

to do another one, it's another 5 to 10

minutes. What we do instead is we

parallelize it. Basically, we just do

all of that simultaneously. So, we come

up with 100 ideas. For every idea, we

come up with 100 or 200 images. For

every image, we come up with, you know,

100 or 200 videos. And we do all that in

the same amount of time as it would have

taken just to do one.

Because Codex and GPT-5.6 Soul make use

of sub-agents really, really

effectively, you can legitimately run

this entire thing in like 10 minutes and

make a million ads you know for whatever

uses that you're going to be using for

the next like 3 months. That's probably

not necessarily a great idea because I

bet you that all of all of this will get

even better by then and you can do even

cooler things, but just wanted to give

you guys a quick example. Okay, so what

I'm going to do is I'm just going to

grab the like markdown file here and

then I'm just going to copy and paste

this into a Ghost TTY terminal.

Paste that in.

And then now it's just going to run. And

the whole idea here is you know it sets

a goal for itself based off of this and

then it will run. And the reason why I

always like having this set up in like a

six panel way is just because this

allows me to do other things while it is

running. So this will this will set

everything up. This will hook to the

website. This will publish the website.

It'll even pick an idea. It'll just

serve you the link basically at the end

which is really sweet. Obviously as

mentioned this is going to consume

credits which you're going to cost. This

is going to consume

GPT image credits which you're going to

cost. This is also going to consume your

usage if you are on the the chat plan,

but you know AI is free these days and

that's kind of how it works. So while

we're at it, let me show you the outputs

that we're getting inside of Pixelfield.

I'm just going to go back to this little

video panel and then we're going to take

a look at what's going on. So as you can

see here this is a five second premium

product beauty shot. Lock the exact

bottle geometry, ivory pump, glass

reflections, label design, and a

correctly spelled word clear hour.

Okay, from this frame.

You can see essentially what it's doing

here is it's already generated an image

of this beautiful spray bottle called

clear hour.

And now it's generating a video for

that. If I scroll down I mean we're

doing a lot with Pixelfield right now.

Like we're publishing all eight of these

simultaneously which is pretty cool.

Before we were using Nest Lock which is

like your little I don't know it's like

a key sort of lighter thing. As you

could see we're kind of sticking it in.

I mean you know would I consider this to

be perfect? Like no. You're going to get

some inputs that don't really make sense

like the key isn't going all the way.

But that's what the verification loop is

for. It tends to catch most of these.

This is a man wearing refined vertical

travel men's wear. I mean this looks

pretty sweet right? Like it's pretty

sexy. Um damn, I wish I had those pants

right now. I'm wearing some 501 Levi's.

Uh you can see that we generated

multiple variants of this sort of nest

lock thing, which is kind of neat. Um

shows you like the range of motion,

moves around. This is interesting, it's

called palm turn. Just basically like a

screwdriver that um I I suppose ratchets

really easily. Maybe it's just like a

clean sort of ratchet.

We even have videos for those pants as

shown over here. Looks like it's a slow

rotation, which is neat. Um I'm a little

bit laggy cuz obviously I'm doing a fair

amount right now. Rain roll, it's like

one of these bucket hats except a bucket

hat that is 100% rain resistant. Uh man

doesn't look too happy about it, but

what are you going to do? And then yeah,

I mean, we just have so many of these.

What's worth pointing out is because uh

as mentioned, there's a variety of these

different models in Higgsfield like the

MCP connector is being used across a ton

of different approaches. Like this is

Claim 3.0 Turbo, but if you scroll up,

you'll see that we're also using C-Dats,

too. So, we're actually using like a ton

of different models in different

approaches. And this is what I mean by

just like letting the model do it on its

own. How about this? This is like a

makeup product for like vertical ads

that it just came up with. Um the reason

why it's laggy again is just cuz I have

multiple of these going simultaneously.

That looks clean as hell. I love the

font, I love the design, I love the way

that it's kind of white, you know, a

little padding. Um yeah, yeah, and you

know, we just have this running

completely autonomously right now.

There's no human in the loop. Uh I am

just straight publishing these to a

website. And hopefully you guys see how

easy it would be to weave this into

maybe some sort of like ads-based

workflow. Where in addition to that, we,

you know, publish this on the internet.

Uh I don't know, make content about it,

do anything like that. You guys are

probably going to start seeing a lot of

these ads in the near term. So,

be careful with that, obviously. Make

sure the products that you buy are

genuinely real products, but yeah, the

future advertisers of the future are

using this exact workflow right now.

They're just not making YouTube videos

telling everybody about it.

Okay, so hopefully you guys appreciated

that video. You guys saw how cool and

straightforward and easy it is to do.

What I want you guys to do right now is

um I have all of the prompts and

everything that I used to create this

down below in Maker Zero. It's a free

community that I put up that basically

just hosts all of the resources that I

give out because I don't believe that uh

actual technical AI automation knowledge

should be gated. Uh I'm happy to just

give all that away for free. If you guys

like this sort of thing and you want to

convert this a big product generator

into some sort of monetizable service or

product and sell it to people, um then

check out Maker School. That's going to

be the first link down below in the

description, which is my AI automation

community where I literally walk you

through the monetization. So that

involves like 90-day accountability, a

guarantee where you'll get your very

first paying customer in that time

period or uh you know, you get a full

refund. It includes a big group of

people that are rooting for you,

cheering for you, and showing you what's

possible. And we have so many people

closing big deals for systems like I

just showed you every day. It's probably

probably probably the community with the

highest actual like return on investment

rate in school right now.

So I'll stop tooting my own horn. Uh

thank you very much for your time and I

really appreciate every second you spent

watching this video. Looking forward to

catching all of y'all in the next video.

See you.