10X AI (Issue #30): Pika Labs 1.0, SDXL Turbo, Meal Roasts by AI, and Pillow Snow

PLUS: Meta's Seamless models, DALL-E 3 in MS Paint, Perplexity's Online LLMs, Amazon's AI models, and removing objects in Midjourney images.

Happy Sunday, friends!

Welcome back to 10X AI: a weekly look at beginner-focused AI news, tools, and tips.

Let’s get to it.

This post might get cut off in some email clients. Click here to read it online.

🗞️ AI news

Here are this week’s AI developments.

1. Pika Labs announces V1.0

I must admit, I assumed that Pika Labs’ Discord bot was already version 1.0.

Apparently not!

Pika Labs just announced V1.0 of its text-to-video model, a full-fledged website interface, and a host of new, impressive features:

Here’s some of what we’ll soon be able to do:

• Apply diverse video styles (3D animation, anime, cartoon, cinematic, etc.)

• Expand the video canvas on the fly

• Replace objects in the video (Adobe only teased this in Project Fast Fill)

And we won’t have to depend on Discord to do any of it.

I’m on the waitlist, and you can sign up as well.

2. Stability AI launches SDXL Turbo

Hold on to your top hats, because Stability AI just dropped SDXL Turbo, a text-to-image model that responds to prompts in real time.

I tested it out, and it works as advertised:

The model weights are available on Hugging Face.

If you want to take it for a spin like I did, head to Clipdrop. You get a few free credits before having to upgrade.

3. Meta’s expressive translation models

Last week, ElevenLabs released speech-to-speech that transforms your voice while preserving its inflection, pace, and so on.

Well, Meta’s Seamless models are like that, but for translation.

Here’s me going full ASMR in English:

And here’s the Spanish version:

Here’s me doing…I honestly don’t know what:

And here’s the same nonsense in French:

You can visit Hugging Face to check out the models or try the Meta demo (which is what I used above).

4. Perplexity releases two online LLMs

Perplexity just launched pplx-7b-online and pplx-70b-online, two LLMs that incorporate fresh knowledge from the Internet in their answers.

This makes them better at responding to time-sensitive queries and more factual than several comparable models:

Source: Perplexity

In my test, pplx-7b-online took just a second to give me an accurate answer:

You can take both models for a spin over at Perplexity Labs.

5. Microsoft incorporates DALL-E 3 into Paint

Announced back in September, Microsoft is finally starting to roll out its “Cocreator” tool, which brings the power of DALL-E 3 directly to MS Paint:

Source: Microsoft

You can generate an image and work on it immediately without having to switch apps.

For now, Cocreator is only available in the US, UK, Australia, France, Canada, Italy, and Germany, so the rest of us will have to wait a bit longer.

6. Amazon finally enters the AI assistant game…

Amazon announced Amazon Q, an AI assistant that’s primarily aimed at businesses.

It’s designed to help employees “have conversations, solve problems, generate content, gain insights, and take action by connecting to [a] company’s information repositories, code, data, and enterprise systems.”

Update 03-12-2023: Days after launch, there are red flags about Amazon Q exhibiting “severe hallucinations” and, more concerningly, leaking confidential data. If true, it’s an early blow for Amazon, which billed Amazon Q as a more secure and reliable AI for businesses compared to offerings like ChatGPT.

7. …and the text-to-image game

The company is also bringing a text-to-image model called Titan Image Generator into its Amazon Bedrock platform.

Here’s the official demo, which proves that Amazon is somehow capable of sucking every ounce of joy out of something as fun as the magic of AI images:

🛠️ AI tools

I always enjoy silly sites that use AI to roast stuff, so I wanted to share another one today.

8. Gordon RamsAI

Edit 18-03-2024: The site is no longer live.

Gordon RamsAI convincingly roasts your home meal in the classic no-mercy style of Gordon Ramsay.

Simply upload a photo of your meal…

Meal image source: The Sun

…and several moments later, you’ll have a rather believable AI Gordon Ramsey making you reevaluate your life choices:

Have fun feeling terrible about your culinary skills.

💡 AI tip

Here’s this week’s tip.

9. Erase objects in Midjourney using Vary (Region)

I previously showed how Vary (Region) can be used to remove unwanted objects in images by prompting it with whatever’s in the background.

But, as it turns out, the process can be even simpler. You can erase objects by submitting an empty prompt.

Here’s an image of a living room:

If I want to remove the picture above the sofa, it’s enough to select it using Vary (Region) and delete the original prompt, leaving the prompt space blank:

This consistently generates results without the selected object (although the bottom-right image ends up with an odd shadow effect):

There you go: A quick way to erase stuff in images without leaving Midjourney.

🤦‍♂️ 10. AI fail of the week

Snowflakes shredded pillows falling on a car

Sunday poll time

---

Previous issue of 10X AI: