Happy Thursday.
Miles, who runs a design studio, posted a launch video he made with Opus 5.5 and wrote, "This video took 3 prompts."
We'd asked for motion and video workflows for today's Dive Radio, and many of you delivered.
Motion used to be the skill you hired out. Now anyone can make things move. What's left is deciding what should move and why. Let’s dig into it.
– Tommy (@designertom)
Open to everyone until Friday, October 9
The people sending us these workflows should be able to find each other, so Ridd and I are opening the Dive Discord to everyone.
While you're there, send us your portfolio in the Discord, with one specific thing you want feedback on, and we'll review a batch live next Thursday, Oct 8.
What should move, and why
Multiple designers shared their workflows with us in videos for today’s livestream.
Bruno Reis had an agent edit a long interview inside DaVinci Resolve. Resolve's API can't split a clip, so they worked around it: drop a one-frame dummy clip where the cut goes, then delete it. "It's kind of funky but it works."
Then the agent made a second, "ruthless" version that cut everything it didn't think was needed. That's how Bruno edits long interviews himself. "It kind of picked up on this habit, which is crazy."
Rebekah Farr built a shop for exotic alien plants where every screen is generated video. Click a plant and a layer in the middle guesses what the click should do, then asks a fast video model to draw the next state: price, description, how many are in stock. The real data sits on the back end. She calls it "very proof of concept."
George Hastings went the other way. His earlier popup for Unicorn Studio had a complex hover animation, and in the first day or two "like 90 something percent" of people dismissed it. So he tried a new announcement built as shaders and code he could art direct "rather than just prompting and waiting on a video." Your mouse moves the light.
The moving part came cheap. Stefan Poulos, making a promo for his portfolio, found the work was somewhere else: "I had to tell it to show restraint."
When a launch video takes three prompts, your job is deciding what shouldn't move.
Look at the motion in your own product, the onboarding animation or the hover state on the pricing page. Did it come from a decision about the person using it, or could a model make it in three prompts? The second kind is about to stop being worth much on its own.
Where generated video belongs inside a real product, nobody knows yet. Rebekah's visuals drift. George isn't sure "interactive cinematic in-product experiences" will "be a thing." He also said "it sure is fun to build."
Ridd and I go live today at 1 PM Pacific.
The main segment is the workflows, including Andy Orsow on drafting AI video with cheap models before paying for the good one, and Nim, whose model ended up animating in Blender, which he hadn't planned.
First, my DevDay picks: dots, the Decisions API (OpenAI's version of last week's Jev), Sign in with ChatGPT, and plugins. Then a short look at America.gov, which puts 29,000 government websites behind one interface.
Models change every week. Jobs change slower than that, and we would rather measure them than guess. If you do design work, tell us what it looks like right now.
With your help, we’ll be publishing the open report this November.
