David (00:02)
Piece of trash. View trash. Is is Omni in there?

Hey Ilan, so ⁓ Google released Omni. ⁓ let's go give it a try today.

Ilan (00:15)
Yeah, let's do that.

David (00:29)
All right. So Google Omni is something that Google announced at Google I/O 2026 not too long ago. as Google's new native multimodal AI model ⁓ family, I guess. I guess it's more than one model, I don't know. ⁓ but

You know, what what what matters is that like its ability to generate awesome video is taking the internet by storm right now.

Ilan (00:53)
What does a multimodal model mean?

David (00:56)
That means that it can understand and probably also generate ⁓ audio and video and image and text, of course. Yeah.

Ilan (01:02)
Okay,

cool. So it's just a one-stop shop model.

David (01:06)
One stop shop. Don't go anywhere else. Don't go to Dario. Don't go to Sam. Just just stay right here with Demis, all right?

Ilan (01:10)
Ha ha.

Don't

load models, local models into Comfy UI.

David (01:17)
The that's right.

Don't even go to yeah, don't even go to Zuck for for Gemma. Yeah.

Ilan (01:22)
Right.

David (01:22)
Alright. Okay. Well, we're gonna find out today. ⁓ so this is the their landing page, and you can see that you can try it directly in Gemini. ⁓ and but today we're gonna try it in Google Flow, which is more of their like video editing app. now before we dive into that, ⁓ it's important to note that there's this learn how to prompt part of it, right? And so let's let's click into that. ⁓ because there's a whole page that ta talks about

Ilan (01:23)
Alright, how does this work,

David (01:50)
How to best prompt Gemini Omni. Okay. ⁓ and so yeah, cool, we can we can read all of this. ⁓ but what I did was is I extracted all of this into a markdown file. And then I I popped it into a Gemini gem. ⁓ so here in Gemini, I've got the Omni Prompt Writer, ⁓ where I'm saying, hey, you're a world class prompt engineer, and by the way, here is the markdown file, ⁓

That basically extracts that page in terms of how to write effective Omni prompts.

Ilan (02:21)
probably a lot of people at this point are doing this, but for anyone who's not, the reality with AI is that you should just be extracting pages, point your LLM to a page, say, get all the content out of this page. Like if all you're doing is going to a page and reading so that you can prompt better differently in another LLM, then there's no reason to do that. Like cut out the middleman.

David (02:45)
Yeah, exactly.

Ilan (02:46)
By the way, the middle man is you.

David (02:50)
Cut out yourself.

Okay, so let's use this gem.

this is where we're going to ⁓ write our prompt. Okay. So if you're curious how we came out with that prompt, it's because we used ⁓ this gem in Gemini.

Alright, so here we are in Google Flow, and this is where we are going to be able to access Gemini Omni ⁓ to generate and maybe remix some videos. Now you do require

having a paid subscription in order to use this new ⁓ model.

Ilan (03:23)
Sounds like there's one more subscription that I'm gonna need to have here.

David (03:28)
Yeah, one more subscription. Just just sub to all all three of the major ⁓ AI providers. You can't go wrong.

Ilan (03:35)
Yeah, that's right. Just a hundred bucks a month.

maybe the next product we should create is the LLM Sports Pack Bundle. know, like when you used to buy from your TV provider, you're like, oh, want ESPN. And they'd be like, okay, to get ESPN, you need to buy all the sports packs. Like we'll do that, but for the LLMs, you want Google? All right, we give you Google, Anthropic, and OpenAI for the low, price of $75 a month. You save 25 bucks.

David (03:59)
And we'll throw in Mistral there too, that's a promo offer.

Ilan (04:01)
That's right. Yeah, exactly. And Cohere.

David (04:04)
And go here, yes. Let's not forget our ⁓ Canadian neighbors.

Okay, so what I'm gonna do is I'm gonna look for some royalty-free videos to tinker around with. So here we are in Pixabay, and we're gonna use this video here. ⁓ thank you, Matthias. and over here we're also going to use this other video for our second one, where it's just an aerial shot of of some nice some nice scenery

Just drop it in. Okay.

Okay, I don't know what's going on here. ⁓ it's it's not a very big file. It's only 88MB.

Ilan (04:43)
all your uploads being used for recording the podcast.

David (04:46)
That's clearly what it is. I'm gonna refresh and see what's going on.

The fact that it's staying at zero percent is concerning.

Ilan (04:51)
Well, it looks like we found a bug in Google Omni. Great job, Google. Way to make products that really serve your users super well.

David (05:00)
This is customer delight right here. I'm so delighted.

Ilan (05:02)
Yeah.

It remixed the video into this sort of like swirling black and white pattern. Give me the avant-garde version of the video.

David (05:08)
That's what you wanted, right? We did the ⁓ everything for you.

I'll try a different way. I'll try it maybe over here. Can I add assets here? Upload media. Let's try that.

Okay. Uploading video. Maybe it works here.

Ilan (05:24)
same swirling pattern.

Well, call me underwhelmed.

David (05:27)
Ha ha ha.

This is a disappointment, Google. I mean they you know the the URL here is labs.google. So I mean they did hedge the naming to say, well, it's experimental.

Okay. I'm gonna try it with a different a different file. Maybe it's the file that's the problem.

Okay, well the clicking on the new project and adding an asset didn't seem to work. ⁓ maybe somebody at Google was vibe coding an an enhancement and they just checked the straight into prod.

⁓ there's ⁓ try Omni Now button over here. So let's click on that and see where what we get. Okay, so I guess it's saying for us to ⁓ upload the video to the to the chatbot. And then

And then we'll take it from there.

Ilan (06:09)
Alright.

holding my breath.

David (06:11)
Yeah.

Ilan (06:12)
Well now I can understand why people can't figure this out with comfy UI. Google can't figure it out. Some guy with vibe coding in his basement probably can't figure it out either.

David (06:18)
If Google can't figure this out, nobody can.

Man, man. That's disappointing. Okay.

Ilan (06:27)
How big of a team do

you think they have at Google working on this to blow it this hard?

David (06:32)
Probably like like 12 to 24 AI agents.

Ilan (06:36)
Hahaha

David (06:38)
Alright, well uploading a video doesn't seem to work today. ⁓ so let's try uploading this image and see whether we can just get that animated. How about that?

Okay. All right. Well we were able to to get an image uploaded. Congrats, Google. You're now in the era of 2024. Okay. So ⁓ let's let's generate a prompt ⁓ for this person ⁓ walking, I

Ilan (06:49)
Hallelujah.

David (07:02)
Let's go.

Ilan (07:03)
I like that it called its prompts masterfully crafted.

David (07:06)
D do you not call your own work masterfully crafted? I do that all the time.

Ilan (07:10)
Every time my boss is like, hey, could you give me an update on where that feature is? This is a masterfully crafted executive ready...

David (07:20)
Ha ha.

Ilan (07:21)
PowerPoint slide?

David (07:24)
Y you

need so you need to open with of course I can. Let me get on that right away.

Ilan (07:27)
You

with three beats first, you know? Of course. Great question, boss.

David (07:28)
Okay. Yeah, exactly.

What an excellent request. Okay, let's let's get started. Here's the here's the generated prompt. It's it's quite lengthy, I'm not gonna read it all out, but we're gonna give it this image, give it this prompt, and see what we get.

we have a failed generation and I suspect that's because we name somebody whose style we probably shouldn't explicitly call out here. So let's try that again without that ⁓ call out.

Ilan (07:51)
Mmm.

David (07:58)
Okay, we see some progress here. Greater than zero percent. I think that's that's a good sign.

Ilan (08:04)
that

is progress.

David (08:05)
Google Omni greater than zero percent.

Alright, well we have our videos and I suspect that it did not take the original image much into account. ⁓ I mean the dog kind of stayed the the same breed. But let's let's see what we got here.

Okay. It's it's like cartoonish, that one. Let's see about this one here.

Ilan (08:23)
Mmm.

David (08:37)
Well, it got the colors all right. Yeah. there were some artifacts of AI generation there. You saw you saw some things get a little bit blurry or not get a little bit chopped. ⁓ but like I don't like how it it you know didn't adhere to the the person here.

Ilan (08:50)
Mm-hmm.

David (08:56)
All right. So ⁓ let this be a lesson in ⁓ reading the the prompts, reading the AI output. ⁓ so I've I've hand ⁓ changed ⁓ some aspects of this prompt ⁓ so that it doesn't overwhelm the original image. ⁓ so previously that prompt it demanded like specific things that that we don't want. ⁓

Like like s you know, certain pastel colors, which, you know, is very Wes Anderson style. ⁓ but I I wanna keep it more close to to this you know, more realistic scenery. So let's have a go.

All right. Okay, cool. ⁓ it completely ignored my request to to to keep the identity of the girl the same. ⁓

Ilan (09:37)
Do think

that it might have used a mixture of your prompt and its previously generated videos?

David (09:42)
I I suspect it it it might have, ⁓ but it it should have remembered the original image here. So that that's really strange to me. ⁓ but let's see what we got.

Ilan (10:01)
I would say We've gone the wrong direction.

David (10:04)
Ha ha ha.

Ilan (10:05)
Omni, you're getting colder.

David (10:07)
You're getting colder, yeah.

Well, it's very cartoony. ⁓ so maybe that's ⁓ that's an issue with the prompt that I that I gave it. but ⁓ I I it's it's too bad that it wasn't able to to ingest a video. ⁓ because like it would have been nice to to throw in a video and then modify the style of that video. I think that you know, taking an image and making a video, that's like a little bit blasé these days. ⁓

But like to to modify a video, like change its style entirely, add a thing to it, that is ⁓ closer to the frontier of what's possible.

Ilan (10:57)
Yeah, my first impressions of Omni, which I had not played around with before we hopped on this recording, are can't upload videos, image to video generation doesn't work very well, ignores your prompts, I'm, color me unimpressed.

Well, we got somewhere. Not sure it was where we expected to be, but hey, Omni's a new product. I'm sure in a few weeks the rough edges will be sanded down and it'll be working great.

And in the meantime, if you want to see more great content on how to use AI products, then follow prompt and circumstance.

David (11:32)
All right, we'll see you next time.

Ilan (11:33)
See you then.