Grok Can Now Generate Videos, With a “Spicy Mode” Included 🔥

Welcome to another edition of Horizon AI,

Elon Musk’s xAI teased video generation for Grok last week. Now, it’s finally here as a new feature called "Imagine."

Let’s jump into it!

Read Time: 4.5 min

Here's what's new today in the Horizon AI

• xAI’s Grok Gets Video Generator Feature, With a "Spicy Mode"

• Anthropic Proposes "Persona Vectors" to Steer LLM Behavior

• AI Tutorial: Instantly swap people and objects in any video using AI

• AI Tools to check out

• AI Findings/Resources

• The Latest in AI and Tech 💡

AI News

XAI

xAI’s Grok Gets Video Generator Feature, With a "Spicy Mode"

The new feature is called Imagine, and it allows users to create both images and videos from text prompts, as well as turn static images into videos

Details:

• The model can generate images in seconds and continues auto-generating new ones as you scroll. These can then be turned into videos, or you can upload your own images to use instead.

• The videos are 15 seconds long and feature native audio. They’re also generated much faster than other tools, with Musk claiming it takes “1/2 to 1/4” the time of major competitors.

• It includes a “spicy mode” that allows users to generate sexually explicit content, including partial nudity (although there are limits to how explicit it can get).

• It is available to SuperGrok and Premium+ subscribers.

The NSFW content doesn’t come as a surprise from xAI, especially after it recently released a raunchy anime AI companion just a few weeks ago. Regardless, the feature is quite impressive, and Musk claims the model will “get better every day.”

Source

TOGETHER WITH SHUTTERSTOCK

Training cutting edge AI? Unlock the data advantage today.

If you’re building or fine-tuning generative AI models, this guide is your shortcut to smarter AI model training. Learn how Shutterstock’s multimodal datasets—grounded in measurable user behavior—can help you reduce legal risk, boost creative diversity, and improve model reliability.

Inside, you’ll uncover why scraped data and aesthetic proxies often fall short—and how to use clustering methods and semantic evaluation to refine your dataset and your outputs. Designed for AI leaders, product teams, and ML engineers, this guide walks through how to identify refinement-worthy data, align with generative preferences, and validate progress with confidence.

Whether you're optimizing alignment, output quality, or time-to-value, this playbook gives you a data advantage. Download the guide and train your models with data built for performance.

Download the guide

ANTHROPIC

Anthropic Proposes "Persona Vectors" to Steer LLM Behavior

Anthropic just introduced a new technique for monitoring, controlling, and even preventing specific personality traits in language models like evil, sycophancy, or hallucinations.

Details:

• Anthropic’s researchers identified patterns of activity within an AI model’s neural network that control its character traits, which they called persona vectors.

• They compare these to parts of the human brain that “light up” when a person experiences different moods or attitudes—similar to how large language models sometimes develop unpredictable personalities or behave in strange or even “evil” ways.

• The researchers propose that these persona vectors could be used to monitor and control a model’s character, as well as identify training data that might lead to undesirable behavior.

As AI systems become more advanced and integrated into daily life, this kind of research is a valuable step toward keeping models aligned and under control. That said, there’s still a long way to go in understanding the inner workings of AI.

Source

AI Tutorial

Instantly swap people and objects in any video using AI

Pika Labs' new PikaSwaps lets users seamlessly swap objects and characters in real-world videos with entirely new elements. Here’s how:

1. Go to the Pika Labs platform and select PikaSwaps from the available AI tools.

2. Upload your video. (PikaSwaps currently processes up to the first 5 seconds of a video.)

3. Select the object to modify:

• Use automatic object selection by simply typing the name of the object you wish to modify.

• Alternatively, use the brush tool to manually select an area for modification. (You can adjust the brush size for better accuracy.)

4. Choose your modification method

• Text Prompt: Type a description of what you want to replace the object with.

• Reference Image: Upload an image that closely matches the new object for better AI accuracy.

5. Click Generate, and the AI will seamlessly replace the object in your video. If you're not satisfied, use the Reprompt or Retry options to fine-tune your results.

6. Download the video or share it on social media.

AI Tools to check out

🔥 Cubox: AI read-it-later App. Save once. Know forever.

✨ Ideogram Character: Character consistency model that works with just one reference image. Render infinite variations of your characters with striking fidelity.

📱 Droidrun: It lets you create and control autonomous agents across native mobile apps, perfect for testing, automation, and experimentation.

👉 Inkr: Transcribe everything — fast, accurate, and seamless.

💥 Macaly: Build apps by just describing them.

AI Findings/Resources

👉 Anthropic- Prompting 101

👀 Why I believe in AGI (again)

🔨 A hiker was missing for nearly a year. Then an AI system spotted his helmet

💡 Anthropic- Vibe coding in prod

The latest in AI and Tech

Alexa+ to Insert Ads Into Conversations

Amazon CEO Andy Jassy told shareholders on Thursday that there’s “significant financial opportunity” in delivering ads through Alexa+, the company’s new AI-powered voice assistant.

ChatGPT is on track to reach 700M weekly users

According to OpenAI, ChatGPT is nearing 700 million weekly active users—up from 500 million in March and four times higher than a year ago. The company has also seen growth in subscribers, now reporting 5 million paying business users, up from 3 million in June.

Perplexity accused of scraping websites that explicitly block AI access

Cloudflare published research that allegedly proves the AI startup scrapes content from websites that have explicitly disallowed it, while trying to hide its tracks.

Apple working on 'ChatGPT-Like Search'

Apple is hiring engineers for an "Answers, Knowledge, and Information" team that’s working on an “answer engine” capable of responding to questions using information from across the web. It’s unclear whether this will be a standalone app or a feature integrated into Siri, Safari, or other Apple products.

That’s a wrap!

Thanks for sticking with us to the end! Let’s stay connected on LinkedIn and Twitter.

Not subscribed yet? Sign up here and send it to a colleague or friend!

See you in our next edition!

Gina 👩🏻‍💻