Google Wants To Flood Your Feed With AI Slop

Recently I’ve been seeing more and more AI slop ads. Can’t you pay for some stock footage or something? Like at Eurovision this year they just had some of the worst AI slop. Is the European Broadcasting Union so broke that they have to resort to AI slop?

But Google has a solution for all your AI Slop woes: just make the AI slop less sloppy. Yes, the headline news from Google IO this year was their new video generation system: Gemini Omni. Technically Gemini Omni Flash, apparently there will be a Pro version launching soon. What happened to Veo (their previous video creator system)? ¯_(ツ)_/¯ I guess everything is Gemini now.

Well, actually, there are some changes from Veo. Gemini Omni is a world model. That means it has a sense of how the world works. Meaning accurate physics and lighting. It’s also much smarter and hopefully will have fewer objects popping in and out of existence.

[embed]

It also allows you to replace objects. And this I think might be an even bigger deal than just video generation.

Just look at some of the demos on the official Gemini Omni page:

The power of the sun in the palm of my hand

It reminds me of an Apple Ad where they were replacing all the things in the video with Final Cut Pro tools. I think it was the Final Cut Pro 11 ad, but Apple decided to take it down for whatever reason and now I can’t find it anymore. Thanks Apple.

Well, now you don’t need Final Cut Pro or any video editing skills to replace things in a video. You just need to ask Gemini.

There were also two other demos I was particularly surprised by. The first is this video of a science explainer.

I mean, the science explainer is a bit light on the details. But you can see how this could be incredibly disruptive to YouTube. I suspect many YouTube channels are already using AI in sneaky ways, but this will supercharge it.

The second was this one where a character (given by a static image) was animated using a video.

This is pretty incredible because previously you would need a mocap rig and a character model to do this. I think some people figured out how to do it with the FaceID sensor on the iPhone but you still need a character model. Now anyone can do this. This is going to be very disruptive.

Although, as usual with these AI video generation tools, you are limited in the duration your clips can be. In this case it looks like Gemini Omni videos are limited to 10 seconds. So maybe you’re not going to make a movie with it just yet. Unless you are OK with cutting every 10 seconds.

Although maybe it could still work. To access Gemini Omni you use a new tool called Flow. I’ve used this tool before to create some header images for my posts. It is very powerful in what you can do, although I only use text prompting.

The tool allows you to upload images, videos, and audio to create projects. You then use text prompts to create things you want to make. It even allows you to specify different characters which will keep your AI slop consistent between the 10 second scenes.

Although one thing you should know about Omni is that unlike previous Google AI tools, this one you have to pay for. Yes, this is another big theme in Google IO. A lot of features are being locked behind the AI Plus or AI Pro plans. Some are even locked behind the AI Ultra plan. Although I suspect this is temporary while they roll the tool out. In the case of Omni it looks like any AI plan will allow you to use it.

Although it looks like you can also use it through… YouTube Shorts?

Today, we’re launching the first model in the Omni family — Gemini Omni Flash. Gemini Omni Flash is rolling out today to all Google AI Plus, Pro and Ultra subscribers globally through the Gemini app and Google Flow. It’s also rolling out at no cost to users on YouTube Shorts and YouTube Create App starting this week.

So it’s very clear what Google is trying to do. They want to copy OpenAI’s Sora. Sora allowed anyone to create short TikTok-style AI slop videos. OpenAI actually discontinued Sora, probably due to being unable to monetize it. But I guess Google thinks they can make it work. Their TPUs are very efficient and they’ve already monetized shorts successfully.

The bad thing about Sora, though, was that people just kept sharing the Sora AI slop to other platforms. Sometimes it even still had the Sora watermark.

I suspect Google will try to label Sora-generated videos. They announced all Gemini Omni videos will use their SynthID watermark which will probably stop the majority of people trying to pass off AI generated content as real. But some people will get around it and as Gemini Omni is a world model, it’s not going to be easy to tell. The same random morphing/disappearing detection tricks are not going to work anymore.

Well, get ready. If you use YouTube Shorts, it’s about to be flooded with AI slop. Well, I mean, more AI slop than usual.