Write a script from scratch, fix one that isn't working, or spin the same idea into a dozen tones.

We put GPT Image 2.0 through 12 real advertising tasks: layouts, product accuracy, style transforms, multi-asset composites, and more. Here's what it nails, where it struggles, and the workflows behind each.

Your AI has processed more than you'll read in a lifetime, and it's bored. This month, create an ad selling it a hobby, class, or activity. Make it persuasive, funny, and make the AI feel seen. Submit a static ad, a 15 to 30 second video concept, or a carousel by 25 May.

Composition is the most underestimated tool in filmmaking. Where you place the camera doesn't just change how a scene looks. It changes what the audience feels, what they notice, and what story they think they're watching. This module takes one scenario, a man trying to put petrol in a Tesla at a desert gas station, and shoots it seven ways to show how much a frame can do without changing a thing about the scene.

Building with AI means holding two truths: data centers are among the fastest-growing consumers of energy on earth, and AI used with intent can displace the flights, sets, and reshoots that traditional production once required. An Earth Day reflection on power, planet, and the discipline of making what's worth making.

Testing lighting on a real shoot costs a day and a gaffer's rate before you have a single usable frame, so most teams skip it and guess. This module runs one scene, a late-night gas station with two people and some tension, through six lighting conditions in parallel: midday sun, dusk, golden hour, overcast, night, and storm light. Same scene, six different stories. It also includes a discovery workflow that renders nine lighting options in a grid, so you see the mood before you commit. See first, decide second.

Overseeing two very different briefs at once, a 30-second video ad for a dental brand and an Instagram Stories series for a beer brand, meant two teams and two ways of working. Using Pencil's new Canvas and a custom Canvas Architect Agent, each project got a board built around how its team actually needed to work: speed and clear ownership for one, feedback loops and quality control for the other. Here's how both came together, plus the agent ID to try it yourself.