HomeVideos

Supercharging OpenClaw With Gemini 3.1 Pro

Now Playing

Supercharging OpenClaw With Gemini 3.1 Pro

Transcript

303 segments

0:00

I'm going to be supercharging my open-claw agents with the

0:03

newly released Gemini 3.1 Pro.

0:06

I was able to work with this model for about four hours

0:08

already before I even made this video and I am absolutely

0:12

blown away by how good Gemini 3.1 Pro is and I'm excited to

0:17

integrate this new model with my open-claw agents because

0:19

previously I was using Gemini 3 Pro and the jump from

0:24

Gemini 3 Pro to Gemini 3.1 Pro is huge.

0:27

If you look at the Arc AGI benchmark, look at this,

0:30

31.1% to 77.1% on Arc AGI 2.

0:35

That is a benchmark related to abstract reasoning puzzles.

0:40

So that is a good benchmark to look to at raw intelligence.

0:43

Also, humanity's last exam, this is academic reasoning,

0:47

same thing.

0:48

Huge jump in improvement from Gemini 3 Pro to Gemini 3.1

0:52

Pro.

0:52

And what you're going to see also is that Gemini 3.1 Pro

0:55

outperforms Opus 4.6 and GPT 5.2 in a lot of these areas as

1:00

well.

1:00

So one thing I want to highlight is I was not expecting the

1:03

model to be this good.

1:04

So when we got Gemini 3.1 Pro and I started using it,

1:07

I immediately can tell when a model is good and I will say

1:10

this, this model is good.

1:13

Now,

1:13

there's only one way that we're going to put that to the

1:15

test in this video and I want to show you guys the current

1:18

state of my open-claw setup.

1:20

So you can see that I have Dario, Elon,

1:22

and Sam who are all configured in my Mac minis and you can

1:26

see here that I now have open-claw configured with Gemini 3

1:30

Pro 3.1 Pro in this open-claw box.

1:33

You can see Sam is currently using Gemini 3.1 Pro preview.

1:37

And something I want to show you guys is what I actually

1:39

created earlier today while on stream for my series of vibe

1:43

coding an app until I make a million dollars.

1:45

Check what I created with Gemini 3.1 Pro.

1:48

Watch this video.

2:33

All right, so that's what I created.

2:34

That is what I created on stream today.

2:37

And in order to create this,

2:38

I actually had to work for quite a bit with a cloud code

2:43

and then I had to use Suno.

2:45

And it took me about maybe like 30 minutes to create that

2:48

video.

2:48

And there was a lot of hands-on stuff that I had to do.

2:51

For example,

2:52

I had to create that and then I had to go into iMovie and I

2:55

had to actually create that soundtrack and then upload it.

2:59

You can see here that I had to upload this and create the

3:02

video and then put the soundtrack behind it.

3:05

And what I want to see is I want to test OpenClaw with

3:07

Gemini 3.1 by making it so that Sam is able to use Gemini 3

3:12

.1 Pro to orchestrate the creation of a complete marketing

3:16

video using a combination of Remotion and then the Minimax

3:21

API to be able to create the marketing video and then

3:24

automatically be able to integrate with the Minimax API to

3:28

be able to add that audio overlay.

3:30

And then in the future,

3:31

rather than me having to have any manual integration where

3:34

I have to use, let's say,

3:36

Remotion and prompt it multiple times and then use iMovie

3:40

to then create a soundtrack over in Suno and then download

3:43

it and upload it via iMovie and then download iMovie and

3:47

then upload that to X,

3:48

I want to see if I can completely automate this process

3:51

using my OpenClaw bot,

3:53

using Sam who is configured with the newly released Gemini 3

3:57

.1 Pro.

3:58

In order to prompt Sam, it's going to be very simple.

4:01

I'm going to be using a speech to text tool called

4:04

BridgeVoice,

4:04

which is one of the products in the vibe coding suite that

4:08

we offer in our BridgeMind Pro plan.

4:10

So I'm going to be using BridgeVoice to streamline my

4:12

consciousness and tell Sam exactly what I want him to do.

4:16

So you guys can listen to my prompt now.

4:18

I want you to build out Remotion and Minimax API together

4:23

so that we are able to create videos that are marketing

4:27

videos for BridgeMind and these videos will need to be 30

4:31

seconds long with upbeat tech music that is instrumental

4:36

behind each video.

4:38

So you can see that the transcription time is essentially

4:41

perfect.

4:42

So the first one is going to be for bridge code.

4:45

Number two is going to be for bridge MCP.

4:48

Number three is going to be for bridge space and number

4:51

four is going to be for BridgeVoice.

4:55

So each video needs to be 30 seconds long and you are going

5:00

to need to build out the system and to build out the

5:04

functionality so that Minimax and Remotion are able to work

5:08

together to create these marketing videos.

5:11

I want you to use the Gemini CLI and launch as many Gemini

5:15

CLI instances to be able to do this for you.

5:18

So if that's going to be the first step,

5:20

what we're going to do next after that will come but let's

5:23

just submit this prompt and we'll let it build.

5:27

In order to learn more about each of these products,

5:29

you can go to the bridgemind.ai website and check out each

5:33

product page accordingly to learn everything that you need

5:36

to know.

5:37

And I'm just going to drop in, yeah, bridgemind.ai website.

5:41

So perfect.

5:41

So I'm going to send this off and we'll see what Sam comes

5:44

back with.

5:44

You can see he kind of like says, hey,

5:46

I saw it but let's see what he comes back with,

5:48

give him some time to respond but he's going to build this

5:50

out for us.

5:51

Okay, so Sam just responded very quickly.

5:53

So you can see I sent that at 440 and at 441,

5:57

he came back to me.

5:58

So he said,

5:59

here's my plan to build the automated video and music

6:01

generation system for the four products.

6:03

So it says we will build a custom node.js pipeline that

6:05

orchestrates the entire process.

6:07

There's going to be a music generation module,

6:09

a remotion video templates module,

6:11

the orchestrator which that's very interesting.

6:15

And then there's going to be, it says based on the, I mean,

6:17

think about this, in under a minute,

6:19

it was able to go to the website and you can even see that

6:22

it picked up on the color scheme of each of these products.

6:26

In less than 60 seconds, it went to the website,

6:29

learned everything there needed to be known about each

6:32

product and then created a plan for the system that we're

6:36

going to build out.

6:37

So you can see this here.

6:39

We're kind of going to just let Sam run with it and we're

6:42

just going to have him build this out.

6:44

So I'm just literally going to add him and I'm going to

6:46

say, okay, execute the plan.

6:48

And that's all I'm going to say, execute the plan.

6:50

And again,

6:51

Bridge Voice is included in the Bridge Mind Pro plan.

6:54

You can get it for 50% off for your first three months

6:57

today for only $10 a month.

6:59

And what's also included in that plan is the Bridge Mind

7:01

MCP, Bridge Space,

7:03

Bridge Voice and soon to be launched Bridge Code.

7:06

But let's give Sam some time because he's now going to work

7:08

on my isolated Mac mini and be able to create the system

7:12

for us.

7:13

Check this out guys.

7:14

So in two minutes, Sam is already back to us and he says,

7:17

I have set up the complete system in two minutes.

7:21

The program handles orchestration using Node.js,

7:24

making the Axios call to Minimax API to generate 30 seconds

7:27

of instrumental tech music based on the prompt,

7:30

then saves it and uses remotion to bind it all into a

7:32

generated marketing video for each product.

7:35

And then here is what that looks like,

7:36

the location of all of this.

7:38

And then now it's asking me to give it my Minimax API key.

7:41

And one thing that I actually noticed while this was

7:43

running is you guys see this little like reaction here.

7:46

Based on what Sam is doing,

7:48

it will change the reaction of what it's doing.

7:52

So for a second, it changed it to a brain,

7:54

then it changed it to a computer.

7:55

And what it's doing is it's saying, hey,

7:56

right now I'm thinking, right now I'm looking,

7:58

I've seen it, right now I'm working on the computer.

8:00

So very interesting that it does this.

8:02

But with that being said,

8:02

I'm going to pass in my API key and I'm going to let that

8:05

work.

8:05

And then we'll get to testing it just in one minute.

8:08

All right, so just a quick update.

8:09

I did pass in my API key and Sam came back to me and said,

8:13

I have updated the EMV file with your Minimax API key and

8:16

started the generation process for all four videos.

8:18

So these videos are now being created.

8:21

So this does take a little bit of time,

8:23

maybe like five to 10 minutes because it takes some time to

8:26

build these videos.

8:27

But so far, I am thoroughly impressed by the intelligence,

8:32

the speed and the intuition that I'm noticing from Gemini 3

8:36

.1 in OpenClaw.

8:38

Compared with Gemini 3 Pro,

8:40

I think it's faster and I do notice the intelligence.

8:42

I mean, so far, this has not made any mistakes.

8:46

It has been very fast and very spot on and just been very

8:50

intelligent overall.

8:52

So this is definitely going to be my go-to model now for

8:56

OpenClaw.

8:57

Previously, I was using Gemini 3 Pro, but that model,

9:01

I was using it because I was using it with my AI Ultra Plan

9:03

with Google and Gemini models are pretty good,

9:07

but the issue was mainly with the hallucination rate.

9:10

And what I see with Gemini 3.1,

9:11

and you can actually see this on some of the benchmarks,

9:14

Google was able to significantly decrease their

9:16

hallucination rate,

9:17

which means that you're able to rely on the model more.

9:20

If we go to artificial analysis,

9:22

what you can see is that on artificial analysis,

9:25

Gemini 3.1 has a lower hallucination rate than Opus 4.6 and

9:30

GPT 5.2.

9:32

So this is now one of the best models to be using with

9:35

OpenClaw and it's also very, very affordable.

9:40

All right, guys,

9:40

so I had some time to play around with Sam in this new

9:43

system that we built.

9:44

There's definitely some tinkering that I want to do to make

9:46

this better so that the styling of these marketing videos

9:49

is better.

9:51

But check out what it did.

9:52

So basically, it created these marketing videos.

9:54

As I asked, I said, you know, hey,

9:56

generate marketing videos for these products.

9:58

And it did.

9:59

So here's some examples.

10:02

That's very loud.

10:09

All right.

10:10

So like this one was very basic, right?

10:12

So then the next prompt I told it I said, hey, you know,

10:14

I want you to make it a little bit more like have basically

10:17

be unique and show styling and unique components.

10:20

And then here's what it came up with with the second

10:22

edition.

10:47

Okay, all right.

10:49

So one thing I will say is that this is like I want to

10:52

highlight a couple key things because this did take a

10:55

little bit of tinkering even to get it working.

10:57

But now that it is working now,

10:59

I do have a pipeline that I'll be able to improve to be

11:02

able to create marketing videos at scale, right?

11:05

I can just have this Sam agent be able to create marketing

11:08

videos for bridge mind whenever I want and the styling.

11:11

I would say I would give that like a 4 out of 10 honestly

11:14

for the styling of those marketing videos.

11:16

I wasn't super impressed with what it came up with.

11:18

But I know from what we did earlier on stream today that it

11:21

is possible to make it just like really like about me.

11:24

If I pass in an example of what I want or some screenshots

11:28

of what I'm looking for,

11:29

it's going to be able to polish that up.

11:30

But the key like the overarching premise that I'm using

11:33

this to show you guys is that these open-claw agents are

11:37

functioning differently than any other AI system that I've

11:40

worked with.

11:41

You're able to work in an isolated environment and work

11:44

with incredible models like Gemini 3.1 Pro to create

11:48

marketing videos just by asking this agent.

11:50

I mean, even with when these agents were building,

11:53

I was walking around my neighborhood and I was texting my

11:57

agent from my phone on the Discord app.

12:00

Hey,

12:00

could you give me an update on the status because there was

12:03

a little bit of an issue with my API key initially.

12:05

Then once I got that set up, then it produced a video,

12:07

then it had an error, then I had to fix it.

12:09

But now it's all working, right?

12:11

And now all I have to do is basically tell it what I want

12:13

and tell it what I want to change and give it some

12:16

basically some feedback of like, Hey,

12:18

I want it to have this or I want it to do this or I want to

12:20

be unique here and here, right?

12:22

But the overarching system is now built and I built it with

12:25

Gemini 3.1 Pro and now it's just going to take some

12:28

tinkering and some polishing and now this is a pipeline

12:32

that I can use for marketing videos.

12:34

So this is going to fit into my workflow to create

12:39

marketing videos.

12:40

You know,

12:40

I posted one of these earlier on X and it got like 5000

12:42

views.

12:43

So what you guys are going to see is I'm learning how to

12:46

use these Open Claw agents in a meaningful way.

12:49

Right now,

12:50

there are way too many influencers and AI influencers that

12:53

are hyping this up to get views and clicks.

12:55

And what I want to do is I actually want to integrate it

12:58

into BridgeMind in a meaningful way and I want to use the

13:02

technology because that's what I'm passionate about.

13:04

That's what I love.

13:06

So that's what we're going to be doing in this series.

13:08

You're going to see more of me working with this and

13:10

something that I did and I'm going to do a live stream on

13:12

Saturday where I'm going to be going over this.

13:14

But I actually wired in to this BridgeMind agents Discord

13:18

the log streams from AWS and Sentry for the BridgeMind API.

13:25

And I believe that I can set up Elon to be able to

13:29

automatically fix Sentry errors and errors that are

13:33

happening in our API and in our front ends as they occur.

13:36

So for example, if there's a 400 error,

13:38

it will be able to ingest that to figure out where it's

13:41

happening in the code to fix it and then deploy it to

13:44

GitHub without me having any integration.

13:47

So we're going to see if we can set that up on a live

13:48

stream this Saturday.

13:50

But this was just a quick example to show you how you can

13:54

supercharge your open-claw agent with Gemini 3.1 Pro.

13:57

I will say it's fast.

13:59

It's intelligent and I highly recommend that you integrate

14:02

with it because it's also very affordable.

14:04

So that's going to be all for this video.

14:06

If you haven't already liked subscribe or joined the

14:08

discord, make sure you do so.

14:09

And with that being said,

14:10

I will see you guys in the future.

Interactive Summary

The speaker demonstrates how to supercharge OpenClaw agents using the new Gemini 3.1 Pro model, highlighting its significant improvements over Gemini 3 Pro, including a major jump in abstract reasoning and academic reasoning benchmarks, outperforming Opus 4.6 and GPT 5.2 in many areas, and a significantly decreased hallucination rate. The video showcases a practical application where the OpenClaw agent "Sam," configured with Gemini 3.1 Pro, is tasked with automating the creation of marketing videos using Remotion and Minimax API. Sam successfully builds the system and generates videos very quickly, demonstrating the model's intelligence, speed, and intuition. While the initial video styling required refinement, the overall pipeline was established. The speaker also shares future plans to use another agent, "Elon," to automatically fix Sentry errors and API issues for the BridgeMind API, further integrating AI agents into their workflow.

Suggested questions

5 ready-made prompts