Tomorrow Tuesday — August 2026 (Issue #3)
LinkedIn just shipped its new feature: “looks like AI slop” — a button to flag low-quality content made with artificial intelligence.
For those of you looking instead for content that’s curated, thought through and made with love, there’s Tomorrow Tuesday: the SOUTS newsletter about the revolution running through film, video games, music and the creative industries.
We explore what happens when artificial intelligence stops being just a tool and starts changing the way we create and consume stories.
What a Week for Open Models
A couple of weeks ago the AI world got shaken up when Kimi K3 came out — a Chinese, open-weights model comparableto Claude Fable 5. This week another model showed up that’s going to change the state of things even more, but its release flew way more under the radar:
DeepSeek V4 Flash
The first “DeepSeek moment” happened in January 2025. OpenAI had just released GPT O1, an ultra-capable reasoning model that cost a fortune. A couple of weeks later, DeepSeek dropped its answer to it: DeepSeek R1.
It was a model almost at the same level, but at a fraction of the cost. It was a brutal shock for the American AI market — a market with the mindset of: the bigger we make the models and the more money we throw at this problem, the better the results will be.
And then China showed up with this model trained for less than 1% of the cost, with a super efficient architecture and a price for the average user that was virtually free.
This week we’re seeing another DeepSeek moment. The Chinese company just launched a model that in some cases is even better than the dreaded Fable 5 — which, if you remember, was banned by the United States government for being “dangerous.”
The big difference is the price. As an example, an average task with Fable can run you $2.75. DeepSeek costs $0.02 for the same task.
That is: more than a hundred times cheaper.
It’s so efficient that at SOUTS we have it running completely locally on our PC — and it’s not like we have some super wow machine.
Between starting to write this newsletter and actually sending it, I had to stop because a new competitor showed up to join the fight.
Alibaba released Qwen 3.8 Max, its most powerful model to date.
If you feel like you can’t keep up with everything that’s happening, welcome to the club. Since ChatGPT arrived in 2022, the speed of AI’s evolution has done nothing but increase.
Welcome to the commoditization of intelligence.
And in video?
This is a newsletter made for the creative world. And while LLMs have their place, you decide to read us for other reasons. So let’s talk about AI video, where things are not stopping or slowing down either.
A new model came out this week: MiniMax H3. Following the trend I’ve been talking about, it’s a Chinese, open-weights model. We are extremely surprised, in a good way.
Quality comparable to Seedance 2.0, and light enough that it runs on basically any modern computer.
It’s available in two variants:
- A model for Text-to-Video and Image-to-Video, with first and last frame support.
- Another model specialized in Reference-to-Video.
The next tutorial on our YouTube channel is going to show how we hooked it up with Hermes to generate video autonomously and completely for free.
But it’s not all Made In China.
We’re also still waiting on Flux 3, which we talked about in the previous newsletter.
Flux.3 is going to be multimodal. It can generate image, audio, video and action prediction. They’re rolling it out bit by bit. It’s comparable to all the top-tier models and they confirmed it will have an open source version, though it won’t be the biggest model.
I hope that, with these advances from the Chinese models, they’re hurrying up — but hey, we’re waiting for it with a lot of excitement.
There are two paths in this AI universe that look contradictory. On one hand models keep getting bigger and more powerful, on the other hand they keep getting smaller and more powerful.
My prediction: at some point in roughly 12 months, the quality gains between models are going to be so marginal that being the best model will mean almost nothing.
At that point value is going to come from other places. Being able to run on devices the size of a Nokia 1100 will be one of those differentiators.
The American response
In the middle of this wave of Chinese open models, more than 230 companies and organizations signed a letter titled Open Weights and American AI Leadership.
The letter says the United States must avoid premature restrictions on open models.
Arguments in favor: open weights let startups, universities, companies and governments access and train AI more easily; it reduces dependence on a handful of private platforms; it helps democratize technological progress.
Arguments against: it’s harder to control open models being used for malicious causes; they can become a weapon for the state’s ‘enemies’.
Signed: Nvidia, Microsoft, Meta, OpenAI, IBM, AMD, GitHub, Cloudflare, etc.
Did not sign: Anthropic.
At the end of the day it’s all politics.
Long Live Cinema
I don’t know about you, but I’m sick of hearing that cinema is dying.
Because of television. Because of VHS. DVD, Netflix, TikTok, artificial intelligence, Generation Alpha, microdramas… 😴
And yet, this past weekend theaters in the United States and Canada sold 430 million dollars.
It was the highest-grossing weekend in the history of the North American box office.
The main one responsible was Spider-Man: Brand New Day, selling 360 million dollars.
But reading the result as just another Marvel win is thinking small. This year a bunch of completely different movies found different reasons to become events.
The Super Mario Galaxy Movie turned THE video game character into a global cinematic phenomenon all over again.
The Odyssey being the most relevant production of the year.
The Backrooms brought a story and an aesthetic born on YouTube into theaters.
And Obsession, with a minimal budget by Hollywood standards, absolutely cleaning up in horror.
There are plenty of bad takes out there, all on LinkedIn: that we need to fund more influencers, more horror, more sequels, more recognizable IP, blah, blah.
My read is a bit simpler: maaaaaybe we all want to go to the movies, we just want to go see good movies.
Who would have thought that if we make good movies people will want to go see them.
Shocking.
Until next Tuesday
I’ve talked plenty already, so that’s it for this week’s issue.
Keep an eye on our YouTube channel, where we’ll be publishing new videos about how to connect open models to generate quality content.
At SOUTSwe try to always stay on top of what’s happening, and every time there’s something relevant we’ll bring it to you in this newsletter.
See you next Tomorrow Tuesday.
Yves