Hollywood explores AI

AI Blog

Issue #15

New model versions for leading text-to-image AIs

Midjourney and Stable Diffusion have announced – one day apart (!) – new versions of their text-to-image AI. Let’s take a look in order at what upgrades and new features each of them brought.

  • Midjourney Model Version 5.2

    Version 5.2 became available on 23.06.23 and has some very interesting new features.

    • New /shorter command: It allows you to “analyze” your prompt and gives suggestions on which words are probably not contributing anything and which are of primary importance.
    • New High Variation Mode
    • New “Zoom Out” Feature: The upscales have a ‘zoom out’ option at the bottom, which reframes the image.

      • Zoom Out 1.5x and Zoom Out 2x pull the camera outwards and fill in the appropriate details on all sides.
      • Custom Zoom opens a pop up window, which allows you to set the aspect ratio, the precise zoom, and also to change your prompt if you want. To change the aspect ratio put –zoom 1 and change –ar to the desired one.
      • Make Square appears under images that are not already square and if selected, fills in details on both sides of the image to make it square.
  • SDXL 0.9 was announced on 22.06.23 by Stability AI and is available via Clipdrop. This model produces much more detailed images. In addition it allows you to generate variations and reconstructions of existing images. Below is the example provided by Stability AI itself, which you’ve probably seen circulating on Twitter.

Marvel & AI 

On Wednesday, June 21, 2023, the first episode of Marvel’s Secret Invasion series was released. All too quickly, Twitter and the various news sites began to buzz about the opening credits being AI generated.

The opening titles are the result of the work of Method Studios and as they said, they felt that Secret Invasion, with the themes it covers, was a great opportunity to experiment with AI, which was used as a tool by the team that was in charge of the project. In fact, Ali Selim (director & executive producer) said “It felt explorative and inevitable, and exciting, and different.”

Why do we find interest in all this?

Because, more or less, we’ve all heard AI influencers claim that this or that “AI tool” is going to take over Hollywood and change the industry once and for all. In fact, such statements were made this past week on the occasion of the release of the new Midjourney and Stable Diffusion models.

Now we have an actual use case and we see that the reactions were not as positive as some would expect. To be precise, a large portion of the public is not ready to accept that materials for films and series using AI are used “without the human dimension and creativity”, especially when they come from giants like Disney, where the resources exist to hire people for the project in question.

Although the use of AI for this particular project was utilized as experimentation with new tools without anyone losing their jobs, as Method Studio stated in a similar announcement made after the backlash, the plethora of negative comments still relate to how people lost their jobs because of this scene or how they may lose their jobs because of the normalization of AI use in the industry. People’s fear of how AI will be instrumentalized by Hollywood is not unfounded, especially considering the terms on which major studios work with VFX artists.

Where are we going with all of this?

Things aren’t black and white. We’re still in a liminal situation regarding the use of AI gen-art tools and will continue to be until the legislative process gets somewhere and until we come to a general consensus on who is the creator of AI gen-art.


AI Insider 📰

  1. Google brings Duet AI to the sheets. Users who are members of Google Workspace Labs can already take advantage of it. We expect to roll out to the rest of us. It’s also rolling out a new feature for Google Search, which it calls Perspectives, that lets you find information that real people have posted on forums and social media sites, so you can get more personal and human results to specific queries.
  2. H Deepmind, η εταιρεία που φτιάχνει τα ΑΙ προϊόντα της Google, ανακοίνωσε πως ετοιμάζει το Gemini, ένα νεό AI chatbot, το οποίο θα ξεπεράσει το ChatGPT.
  3. It’s about time now that many are blaming OpenAI for not being open. These gripes have become more pronounced since mid-March 2023, when GPT-4 was released. Normally the release of such AI models is accompanied by detailed research papers, explaining how they were created, how they work, what training data they have and so on. in the case of GPT-4 however, we received mere breadcrumbs. From the looks of it, GPT-4 is 8 models all joined together like a Lego tower, rather than one huge model. At least that’s what George Hotz seemed to mean in an interview. Of course, until an official announcement comes out from OpenAI, we can’t be 100% sure.
  4. YouTube plans to use AI to dub video in over 70 languages.
  5. The creators attending this year’s VidCon see AI as a productivity tool, not a competitor.
  6. Stability AI continues to face problems at the corporate level, as the head of research and COO resigned this month.

Learning Bytes 🧐

  1. Harvard encourages students in “Computer Science 50: Introduction to Computer Science (CS50)” to use ChatGPT to solve questions they have.
  2. Remember the case with the New York lawyers who used ChatGPT? They were eventually required to pay a $5,000 fine each for the false information in the legal brief they filed.
  3. It seems that OpenAI is trying to lobby the EU, despite their heavy-handed statements that AI should be legislated and restricted.

Cool Finds 🤯

Robot learns how to do household chores by watching YouTube videos.


Articles we read last week that we found interesting and we recommend them to you:


See you next week! 💚




Social Media Awards



E-volution Awards



Peak Performance Marketing Awards



Effie Awards