Sunday Rundown #106: Grok 4 & Scuba Bear
Sunday Bonus #66: Midjourney Prompt Builder that automates the process.
Heads up: This is the last Sunday Rundown for a while. I'm off with my family for the summer over the next few weeks. Regular rundowns will return in mid-August.
Happy Sunday, friends!
Welcome back to the weekly look at generative AI that covers the following:
Sunday Rundown (free): this week’s AI news + a fun AI fail.
Sunday Bonus (paid): an exclusive segment for my paid subscribers.
In case you missed it, here’s this week’s Thursday deep dive:
Let’s get to it.
🗞️ AI news
Here are this week’s AI developments.
👩💻 AI releases
New stuff you can try right now:
Black Forest Labs launched Kontext Komposer, which lets you restyle images without writing prompts by using polished presets. (Try for free.)
Deepgram released Saga, a Voice OS that lets developers turn spoken ideas into code without having to leave their workflow.
Exists released its AI World Creator, which creates playable 3D worlds from image inputs and text prompts. (Try for free.)
Genspark launched an AI Pods feature that turns any input into an audio podcast (a la NotebookLM’s “Audio Overviews”).
Google news:
Circle to Search now incorporates AI Mode, letting Android users get detailed insights through follow-up questions.
Flow users can now animate input images and make them talk using Veo 3, access to which has also expanded to 140+ countries.
Kling AI dropped KOLORS 2.1, an upgrade to its image model with sharper visuals and more style variety.
Microsoft released Phi-4-mini-flash-reasoning, a small reasoning model for edge devices that runs up to 10x faster than earlier Phi versions.
Mistral dropped Devstral Medium and an upgraded open-source Devstral Small 1.1, two cheap, competitive coding models.
Moonshot AI open-sourced Kimi K2, an agentic reasoning model that can handle coding and tool use.
Moonvalley made its video model Marey public, trained on licensed footage and giving filmmakers greater control over scene details and camera motion.
Perplexity launched its chat-driven, agentic browser Comet to Perplexity Max users. The rest of us can join the waitlist. (Sign up here.)
Reka AI open-sourced Flash 3.1, a fast reasoning model for coding, and also Reka Quant, which can reduce LLM size with almost no performance drop.
xAI released Grok 4, currently the smartest reasoning model based on Artificial Analysis benchmarks. (Try it here—needs a paid plan.)
Zoom now has a Custom AI Companion that connects directly to key business applications like Salesforce, Google Drive, and more.
🔬 AI research
Cool stuff you might get to try one day:
Chinese researchers showcased MemOS, an operating system for LLMs that gives them long-term memory and much better retrieval capabilities.
OpenAI news:
Rumored to be working on an AI-powered browser that incorporates LLMs to change how users search the web.
Leaks suggest an upcoming “Study Together” feature in ChatGPT that likely uses Socratic tutoring to help people understand concepts.
📖 AI resources
Helpful AI tools and stuff that teaches you about AI:
“Anthropic courses” [reference]—a collection of free courses by Anthropic for working with its AI models and tools.
“n8n Workflow Finder [Template Database]”—a super handy tool that finds relevant n8n workflow automation templates based on your needs.
🤦♂️ AI fail of the week
Perfect! No one will ever suspect this adventure photo of me isn’t real.
Send me your AI fail for a chance to be featured in an upcoming Sunday Rundown.
💰Sunday Bonus #66: Get copy-ready Midjourney prompts with my Claude Artifact
Last week’s Smart Prompt Maker was a hit, as was the Image Prompt Cleaner from a few weeks ago.
This week, I’m combining those ideas into a long-overdue Midjourney goodie.
Midjourney Prompt Builder helps you go from “maybe some cyberpunk owl thing I dunno lol” to a complete, clean, copy-ready prompt.
All you need is a free Claude account.
Here’s how it works:
You type in your vague idea for an image.
It asks you tailored questions about the subject, scene, style, etc.
You tweak a few Midjourney-specific settings (optional).
The tool turns your freeform answers into a polished, properly formatted prompt.
You paste it straight into Midjourney and generate your image.
No fiddling with Midjourney parameters. No overlooked scene details.
Here it is in action:
Take the Midjourney Prompt Builder for a spin: