🎥 Impressive new videos with OpenAI Sora

Good morning AInauts!

Welcome to the AI newsletter with the most important news, tools & hacks relating to artificial intelligence.

This week is a little quieter, which is why we have some time today to try out a few more tools in addition to the news

That's what we have in store for you today:

• 🎥 Impressive new videos with OpenAI Sora

• ❌ Beware of deception: What doesn't work with video yet

• 📸 Never again a fashion photo shoot with real models?

• 🏆 Claude Update: Officially better than GPT-4 + Create mind maps

Let's go!

🎥 Impressive new videos with OpenAI Sora

Let's start very gently with some inspiration. Some well-known artists and directors have recently been given access to the hyped OpenAI Sora Model.

The result is a few really cool and a bit weird creations. Here is our current favorite video:

The tenor from the test group: Sora is particularly strong when it comes to making things seem real that are not.

The video by Don Allen Stevenson shows this once again in a very impressive way:

(Sorry for the Chinese dub. But we couldn't find another version of the video on YouTube)

All pretty cool, we think. However, we don't think Sora will be available for us normal people any time soon.

The reason: AI video simply consumes an incredible amount of computing power, and the 20 euro ChatGPT Plus subscription per month doesn't go very far. In other words: it will probably be a while before this is really available "at scale". Until then, it's definitely worth marveling at.

But who knows at the moment. Everything is moving so fast, the chips are getting better, cheaper. Let's see ... we'll stay tuned in any case.

❌ Beware of deception: What doesn't work with video yet

While we're on the subject of video and AI, here's an important announcement.

Posts showing alleged user-generated content (UGC) videos that are said to be 100 % AI-generated are currently going viral online.

Like this one, for example:

Important to know here: No, these videos are not 100 % generated with AI.

The person above, for example, is real and you can also find them here on Fiverr.

Basically just a clone of her voice was made with AI and then via lip syncing, pre-recorded videos with new text look pretty real.

One app that you can use to do this is ArcAds. We are currently testing the app ourselves (detailed review to follow).

The concept in and of itself has enormous potential. But it's nonsense to claim that these videos are generated 100 % with AI. So far, the basic videos have always been real. Therefore, depending on the text, you will also have inappropriate gestures etc.

For example, we have already made HeyGen avatars with voice models out of ourselves, which is really impressive.

The technology is getting better and better, but it's not quite where the Internet would like it to be.

📸 Never again a fashion photo shoot with real models?

We know from feedback from the AInauten community that many people are interested in the topic of AI influencers and AI models. And we have already written a lot about how to create AI influencers and the like.

A topic that is always a bit tricky: How do I get my model to wear special clothes from a real brand, etc.?

Generative AI is always a grab bag here, and even if you can get very close with tools like Fooocus, it's still not 100 %.

Thank goodness a lot of smart people are already working on it! We recently talked about the Outfit Anyone Model from Alibaba.

At least a demo is already available here, where you can dress a model in your own clothes. And the results are really impressive!

As you can see in the example above with the pineapple, the whole thing also works with objects, and the AI then builds clothes from them.

And it also works with anime designs!

Here you can try out the whole thing for free on HuggingFace.

We have also found a website for this, where you can upload your own model for a fee, and not just your own clothes.

This saves the detour via a FaceSwap app. However, the feature is still in beta. In general, it shows the trend of what will soon be possible!

🏆 Claude Update: Officially better than GPT-4 + Create mind maps

Ok, let's move on to the last topic for today. It's Claude once again.

After the Metaprompt was very well received last week and we love Claude (at least until GPT-5 arrives), we're building mindmaps with it today.

First of all: Claude Opus performed better than GPT-4 in a practical test with 70,000 testers!

And if you are using a free chatbot, you should definitely use Claude instead of just ChatGPT.

As a reminder: At https://chat.lmsys.org you can compare all important chatbots directly and use them for free!

Practical tip of the week: Mindmaps with Claude

We always think and learn better when things are visualized.

With Claude, you can build mind maps and flowcharts very quickly and easily. You don't even need a fancy prompt, just something like this:

Erstelle mir eine Mindmap, wie ich am besten Prompts für dich aufbauen sollte. Nutze bitte Mermaid-Code.

The result looks like this:

Then copy the code and paste it on this page. And now you have a mind map!

You can then download these etc. The whole thing works not only with mind maps, but also with flowcharts, Gantt charts, etc.

We like to use it to generate a summary as a mind map at the end of a brainstorming session with Claude. Have fun with it!

P.S.: Of course, the whole thing also works with Microsoft Copilot (there is also a GPT) or ChatGPT etc. But we want to encourage everyone to give Claude a try. 😉

P.P.S.: And just fresh from the press, Amazon is investing heavily in Anthropic - it's the biggest investment in the company's 30-year history to date!

We made it! We're through for the week. Monday we'll continue with lots of news, tools & hacks.

Thank you for being there.

See you soon, your AInauts

Follow us on Twitter & LinkedIn!

Your feedback is essential for us. We read EVERY comment and feedback, just respond to this email. Tell us what was (not) good and what is interesting for YOU.