Texti Newsletter #6

Merry Christmas🎄🎅 🛷 to everybody! This is by far the most beautiful time of the year in the Northern Hemisphere! People are drinking hot wine outside, eating all sorts of delicious and winter exclusive foods! What a beautiful time to be alive!

With this occasion we decided to celebrate and warm your house with a unique discount for subscribers. Use the code MERRYCHRISTMAS23 at checkout and get 50% off next 3 months on any of our products!

The highlights of the week:

  1. Midjourney v6
  2. Hands-on Pika.art

Beta launch of Midjourney v6

Twitter is exploding with Midjourney v6's launch! If you browse twitter just a little, you can see everybody show-off their ultimate skills at generating art with Midjourney v6.

As claimed by the developers itself in their blog article the capabilities of the model are exceeding their own expectations. I don't think anybody was ready to see this coming from an AI model yet, it is still just a noise vector approximation, however with some clever algorithms and the right prompt the results are indeed pretty fascinating.

In my opinion the key difference here, is that you really don't need to use the obnoxious keywords we used to mandatory add to the prompt for better results. Finally no more 4k, ultra-hd, high-res, hyper-realistic, ultra-high-super-duper-puper-muper-grouper-high-resolution-instruction. This noise really is annoying and you focus on the prompt itself and forget the keywords, this really is annoying 😖

In case you were wondering if the model managed to learn how to generate text, the answer is somewhat. As described in their own article, you need to get a little lucky and also make sure to include the desired words in "quotes", this way the algorithm understands which words needs to be properly included and written on an image.

On the example bellow, you can see how to stylize the font in the generated image. This is getting closer to an a photo editor that used to take time and effort, now it's just a matter of seconds. Of course the designers have a place in this world, but their niche is getting narrower day by day.



A modern burger advertisement, with the words “THE BIG ONE”, featuring a delicious huge beef burger, with tomatoes and lettuce in the background


A modern burger advertisement, with the calligraphic words “THE BIG ONE”, featuring a delicious huge beef burger, with tomatoes and lettuce in the background

I you look down further in the blog post Midjourney has posted, you can even instruct the AI copy styles of text or image, or even replace the style of text, that is actually extremely valuable because it allows you to consistently edit the desired output and even combine results it for perfect image. This alone is insane, and creates a high amount of possibilities.

My only concern is: will it use licensed fonts? And if so, how many cases of font copyright infringements will there be.

To my eye this is very impressive, and midjourney is getting now the king of image generation market again! With the text prompting this could really disrupt the graphic designers market, as ad generation becomes way cheaper and affordable. Some go further and even say that this Midjourney 6 is a job killer.

After such a demo and hands-on experience, I really must say. I agree.

But hey this a Christmas edition, and I can't leave you on a negative note, so worry not, people are too lazy to generate images by themselves, select and maintain a style, if you do this on your own, you're most likely a poor entrepreneur.

Designers are still going to have a job, yet their job specific is going to change, responsibilities will include preserving the style and brand of the company. Job perspective is going to be rather analytical than artistic, and I think this really interesting.

Hands on Pika.Art

This week I've got invited to try out Pika Art, I was telling you about in the previous newsletter.

Hint, you've already seen an example of the results produced by it. Just after the Christmas congratulations of this email. Yes that is purely AI generated.

After generating about 50 videos and experimenting with all sort of prompt variations, I can safely assure you that our guess was absolutely correct. Video Generations is literally generating about 72 sequential images one by one, and combining them in a 3seconds long video of 24 FPS.

Is it any good?

It's better than nothing 😅, but the results may vary a lot from case to case, it's closer to a DALL-E 2, level than midjourney v5. Images are not very crisp, details are a little lacking, and artifacts are very frequent. It obviously struggles with fingers, text and for video there's a new dimension to struggle about -- motion.

We tried 9 different prompts to make a wheel rotate, and it failed to do it, see the list bellow

Prompt: car wheel rotating very fast close-up

Probably the best result we had, was the following video, where the wheels are barely visible, and the give you a sense of rotation.

Prompt: ford mustang, moving sideways and smoking rotating rear tires on a track at the sunset. take a 3 quarters angle from the back of the car

Probably the best motion result I've got was making this bunny turn around, although the input was different to the output

Prompt: In an enchanted world, create a high definition, high resolution video capturing the magical essence of a mythical garden. The star of this scene is a tiny, fluffy bunny with sparkling eyes and a twitching nose, full of energy and curiosity. This bunny is not ordinary, its fur shimmers with an array of pastel hues under the sunlight, reflecting the captivating charm of its surrounding.

However other prompts yielded worse and way more unpleasant results. Check out this surfer:

Prompt: cinematic short film, about a giant wave of nazare, portugal, with a surfer riding the wave.

Or this dolphins

Prompt: cinematic film of many dolphins swimming in a blue lagoon

In most of these examples we can clearly see that there are algorithms working on creating sequential images, however motion is still lacking. There's plenty of improvement yet to make, but at the moment this is still an alpha test, rather than a ready product. The output is far from consistent and it really depends on the input and the user skills. I am not the best prompt engineer out there yet, but I'm learning and will guide your thought it.

Let me know if you're experience differs.

That's it folks, see you next week ❤️️️️️️️

Happy Prompting!

Remember to invite your friends to subscribe at https://texti.app/newsletter

{{ address }}
Unsubscribe · Preferences

© 2024 Texti Newsletter