Monthly Archives: September 2026

Runway’s reframing tech melts my brain

AI/ML, Illustration

OMG, it’s like Photoshop’s now-ancient (!) Content-Aware Scale, times 10. Check out the thread below, or look in the album I made to gather its six eye-popping examples.

“Love, Rendered”: Can AI remember it for you wholesale?

AI/ML

Honestly I have some very mixed feelings here: as the son of an elderly dad who’s facing increasing challenges around memory, I have a hard time even watching this trailer, much less the full film. Visualizing old memories, reanimating old friends… Is this a good idea? I have no idea, and emotionally I find I can’t much engage beyond the basic premise. Still, because it’s culturally interesting, and we’ll almost certainly see more explorations like this:

Ethelle and Burt have shared a lifetime together, but as Burt’s memory begins to fade, so too does the story of how their love began. Holding onto one cherished moment that was never photographed or recorded, Ethelle works with clinical experts and her family to recreate their first meeting using generative technology. LOVE, RENDERED is an intimate documentary short about love, memory, aging and the ways new tools may help preserve human connection. Directed by two-time Oscar-nominated filmmaker Liz Garbus and produced by Oscar winner Dan Cogan and Oscar-nominee Darren Aronofsky.

Interactive letter-plumping

Typography

Fun!

Seems there’s a whole Insta feed full of related delights, like this:

Netflix asks, “You wanna get high (dynamic range)?”

AI/ML, Image Science, Photography

Check out DiffHDR, which looks useful for making both captured & AI-generated vids more amenable to rich display and editing:

Most digital videos are stored in 8-bit low dynamic range (LDR) formats, where much of the original high dynamic range (HDR) scene radiance is lost due to saturation and quantization. This loss of highlight and shadow detail precludes mapping accurate luminance to HDR displays and limits meaningful re-exposure in post-production workflows. Although techniques have been proposed to convert LDR images to HDR through dynamic range expansion, they struggle to restore realistic detail in over- and underexposed regions.

To address this, we present DiffHDR, a framework that formulates LDR-to-HDR conversion as a generative radiance inpainting task in the latent space of a video diffusion model. By operating in Log-Gamma color space, DiffHDR leverages spatio-temporal generative priors from a pretrained video diffusion model to synthesize plausible HDR radiance in over- and underexposed regions while recovering the continuous scene radiance.

Our framework further enables controllable LDR-to-HDR video conversion guided by text prompts or reference images. To address the scarcity of paired HDR video data, we develop a pipeline that synthesizes high-quality HDR video training data from static HDRI maps. Extensive experiments demonstrate that DiffHDR significantly outperforms state-of-the-art approaches in radiance fidelity and temporal stability, producing realistic HDR videos with considerable latitude for re-exposure.

Tutorial: Use Astra to drive After Effects

After Effects, AI/ML

As I showed last week (i.e. ~3 million years ago in AI time!), it’s now possible to use AI to choreograph the creation of complex animations using After Effects:

Here’s a solid 8-minute tutorial on how to get up and running:

Summary courtesy of Gemini:

The three-step workflow:

  1. Storyboarding (1:50 – 4:05): Before building, use Astra to generate a storyboard. This ensures visual alignment and prevents inefficient prompting loops. You can also provide reference videos for Astra to analyze and use as stylistic inspiration.
  2. Prompting & Building (4:05 – 6:23): Provide a detailed prompt to Codex describing your desired scenes (beats) and effects. You can use provided templates from the creator’s GitHub repository. The AI will then generate scripts, perform the animation in After Effects, and automatically check its own work before presenting the final project.
  3. Editing (6:23 – 7:52): Finalize your project using follow-up prompts to refine specific graphics, transitions, or audio. The creator emphasizes the importance of human oversight, especially for managing transition speeds between scenes.

A little prototype expression-changer

AI/ML, User Interface

Cool work from Gabi on the Firefly team; see details in the tweet.

Ironically, Adobe shipped interactive expression control six years ago (!) in Neural Filters, and I left Google to return to work on that stuff at Adobe. The quality & generalizability of GANs let us down, however. It’s great to see continued progress now.

Go flat->layered with “Ad Delayer”

AI/ML, Design

I can’t say I grok the name, but whatever: not unlike Canva’s Magic Layers feature, Ad Delayer promises, “The ad you already paid for becomes a working source file again.”

Controlling a generative scene via Intangible 3D

3D, AI/ML

I’m a fan of these guys. Charles is awesome, and my friend Philip (with whom I got my start in his pre-ILM/Pixar days) is leading product design. I’m excited to try out their “semantic scene architecture,” which promises to focus on outcomes rather than inputs:

GPT Astra driving AE & Premiere Pro

After Effects, AI/ML

Desire to know more intensifies…

“Everyone in the world now has a 3D designer at their fingertips”

3D, AI/ML

Insane that 3D worlds can be conjured straight from language, then rendered natively or used as guidance for AI-based rendering. Check out this demo from my former Adobe teammate Tom:

Days of Miracles & Wonder, man—always.

Behind-the-scenes thread:

And yet more magic:

Photoshop combines generative imaging & masking

AI/ML, Photography

It seems I’m on something of a Photoshop kick lately (something something you can take the boy out of West 10, but…), so I’ll mention a couple of interesting-sounding enhancements that have landed in the latest release:

  • Instruct Edit with Masks, powered by Firefly Image 5, understands the full context of your image so you no longer need to precisely mark every edit area yourself. Just describe the precise edits you’re looking for — such as opening closed eyes or placing a hat on a specific person’s head — with simple prompts. The unmasked areas remain untouched so your other important elements, like faces, logos, or brand assets, are protected.
  • Light Adjustment Layer gives you professional-grade lighting controls — Exposure, Contrast, Highlights, Shadows, Whites, and Blacks — in a new non-destructive adjustment layer, so you get Camera Raw-level results without ever leaving your Photoshop layers workflow.
  • Markup lets you visually communicate edits by drawing directly on an image to show the model what you want — select areas to recolor, sketch arrows to indicate position, or brush in rough shapes to suggest new elements — so you can get results that more closely match your vision while reducing the ambiguity of text-only prompts.

Omni Flash: Poolside AI

AI/ML, Google Omni

See, I just take Gemini to the dog park, but my teammate Genevieve took it all the way to the pool on vacation. There are levels to this stuff. 🙂

But seriously: this is a really nice visualization of the new clip-extension feature, among other things. That’s how the scene can run a full 30 seconds.