Sunday Rundown #103: Product Placement & "Theby Cats"
Sunday Bonus #63: My Custom GPT to clean up image prompts.
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.
Let’s get to it.
In case you missed it: This week’s Thursday deep dive:
🗞️ AI news
Here are this week’s AI developments.
👩💻 AI releases
New stuff you can try right now:
Adobe rolled out a Firefly mobile app and incorporated more third-party image and video models into its ecosystem.
Browserbase launched Director.ai, a no-code browser agent you can prompt in plain language to take web actions on your behalf. (Try it for free.)
ElevenLabs added v3 Audio Tags that let you fine-tune how AI voices speak—controlling pauses, pacing, emotion, accents, etc.
Google news:
Gemini app for both Android and iOS got video upload support, so you can share your videos for it to natively parse and analyze.
Gemini 2.5 Flash and Gemini 2.5 Pro are now generally available. The new Gemini 2.5 Flash-Lite is the company’s fastest, most cost-efficient model.
Search Live is out in the US, letting you have voice chats with Gemini and get responses based on real-time search results. (Opt in for AI Mode to test.)
HeyGen launched Product Placement (powered by Avatar IV), which blends your product photo, an AI avatar, and a script to create realistic UGC video ads.
Higgsfield AI launched Canvas, an image editing model ideal for product placement, which lets you inpaint items, swap clothes, or change faces in a photo.
Midjourney launched its long-awaited V1 Video Model, which can animate an existing image and turn it into a 5-second video clip.
MiniMax had a 5-day release spree, shipping lots of new stuff:
MiniMax‑M1‑80k is an open-weight, large-scale, hybrid‑attention reasoning model with function‑calling support and a hyphen-filled product description.
Hailuo 02 video model has best-in-class instruction following and native 1080p resolution, climbing to #2 of the image-to-video leaderboard.
MiniMax Agent is an AI agent with native multimodal understanding for complex, multi-step tasks, including coding and tool use. (Try it for free.)
Hailuo Video Agent (beta) lets you pick a style, describe your idea, and get a polished video in one go without editing or prompting.
Voice Design lets you create any voice in any language and customize its emotions.
🔬 AI research
Cool stuff you might get to try one day:
Adobe is working on an LLM Optimizer that helps monitor, benchmark, and optimize for AI-driven traffic (e.g. LLM searches). (Sign up for updates.)
YouTube will let creators use Veo 3 for their YouTube Shorts video clips later this summer, according to CEO Neal Mohan.
📖 AI resources
Helpful AI tools and stuff that teaches you about AI:
“AI prompt engineering in 2025: What works and what doesn’t” [VIDEO] - Sander Schulhoff on Lenny’s Podcast.
“Software Is Changing (Again)” [VIDEO] - a great keynote by Andrej Karpathy at AI Startup School.
“The OpenAI Files” [REPORT] - a comprehensive critical look at OpenAI’s governance practices, culture, and leadership.
“The OpenAI Podcast” [PODCAST] - a new podcast series of long-form conversation with OpenAI staff.
“Understanding the Impacts of Generative AI Use on Children” [REPORT] - a study of AI use and its impact on children by The Alan Turing Institute.
🤦♂️ AI fail of the week
So true! Shorhaip cats can be very affcctiong.
Send me your AI fail for a chance to be featured in an upcoming Sunday Rundown.
💰 Sunday Bonus #63: Fix your bloated image prompts with “Image Prompt Cleaner”
My crusade against “splatterprompting” is well-documented:
But two years later, I’m still bumping into prompts that read like someone puked up a thesaurus.
So I made Image Prompt Cleaner—a custom GPT that takes your bloated prompt, extracts the essence, hands back a clean version, and explains what’s been removed and why.
What does it do?
Pre-emptively flags terms that might trip up censorship filters in image models (you can choose to “sanitize” the prompt or keep it as is).
Identifies foreign-language prompts (you decide whether to receive the clean prompt in English or the original language).
Removes “noise,” like modifiers that typically have little impact on image quality.
De-duplicates overlapping and repeating descriptors.
Re-arranges the prompt structure to frontload key elements.
Real-world example
Here’s how Image Prompt Cleaner fixes a real, in-the-wild “splatterprompt”:
Ready to fix your prompt bloat?
Take Image Prompt Cleaner for a spin: