AI is fixing — and ruining – our photos

Episode Summary

Executive Summary: The episode explores how smartphone photography is shifting from faithful capture toward opinionated, AI-shaped image-making. Vergecast interviews Halide co-founders about Process Zero, a less-processed iPhone capture mode, and Allison Johnson about testing AI photo edits on family images. The discussion centers on authenticity, memory, control, editing norms, and the future trustworthiness of photos.

Main Topics: Halide’s philosophy and the rise of Process Zero (Priority: 5/5): Halide co-founders explain how smartphone camera apps evolved from simple wrappers to tools that let users opt out of Apple’s computational photography. Process Zero deliberately removes most processing to prioritize a more natural, film-like look and user control. RAW, ProRAW, and the meaning of a photo (Priority: 5/5): The guests distinguish native RAW from Apple ProRAW and argue that a RAW file is not yet a photo because it still needs processing. They frame ProRAW as a midpoint between sensor data and final image output, with Apple already baking in corrections and merges. AI editing as memory editing (Priority: 5/5): Allison Johnson describes using Google Photos AI tools to edit family photos by removing clutter, people, or adding objects like birds and dinosaurs. The experiment shows that AI changes not just the image, but how the moment feels and how future memory may be shaped. Authenticity, imperfection, and photographic taste (Priority: 4/5): A major theme is that people increasingly prefer images that feel authentic even if they are technically worse. The discussion suggests that noise, softness, and imperfection can make photos feel more like lived experience than heavily optimized computational images. Where the line is drawn: acceptable editing vs fakery (Priority: 4/5): The panel debates what edits are ethically acceptable—removing background clutter, boogers, or strangers versus adding birds or dinosaurs. They compare consumer photo norms to journalism and professional photography norms, which are clearer about what edits are allowed. The future of mobile cameras under shrinking hardware (Priority: 4/5): Sebastian argues thinner phones and wearable devices will rely even more on processing because physics limits tiny sensors and lenses. This pushes cameras toward data-driven image construction, potentially reducing the role of optics and increasing the role of AI. Running gear hotline: choosing Ray-Ban Meta vs earbuds (Priority: 2/5): In the hotline segment, the hosts compare Ray-Ban Meta glasses, Shokz OpenRun Pro 2 bone-conduction headphones, and Pixel Buds Pro 2 for running. They recommend the Ray-Bans for casual runs and Shokz for more serious runners, noting fit, weight, battery, and environmental noise concerns.

Key Arguments: Modern smartphone cameras are increasingly about processing decisions, not just light capture; users are reacting against invisible AI choices and want more control. Halide’s Process Zero is popular precisely because it reduces processing, even if the resulting photos are objectively worse by technical metrics. A RAW file is sensor data, not a finished photograph; ProRAW is already partially processed and sits between native RAW and JPEG. People often want photos that match memory and feeling, not strict objective reality, which explains the appeal of softer, less corrected images. AI editing can be useful for removing background clutter or unwanted people, but adding content (like birds or dinosaurs) can make images feel less truthful and more generic. The industry may be heading toward cameras that depend on AI to reconstruct reality from limited hardware, especially as devices get thinner and smaller. Editing norms will likely diverge by context: journalism, professional photography, and casual consumer photo sharing each imply different ethical boundaries. For running, the best audio option depends on context: Ray-Ban Meta glasses are convenient for casual quiet runs, but in-ear or bone-conduction options are better in noisy environments.

Data Points: Halide first code commit: 2014 - Ben says the first code commit for Halide dates to 2014, with real work resuming later. Halide public launch: 2017 - Ben notes Halide came out in 2017, around the iPhone X and Smart HDR era. Year the AI photo experiment took place: about a week - Allison says she carried the Pixel 9 Pro and used AI tools for roughly a week of family photography. Google Photos AI tool usage: multiple edits on a week of photos - Allison used reimagine, remove-background-style edits, and other AI tools on several family images. RAW captures merged by Apple: 9 photos - The Halide co-founders describe Apple as taking a burst of nine photos and merging them for Smart HDR-style output. Photo editing threshold for not-a-photo: 50% or more of pixels - Sebastian suggests that if a majority of pixels are filled in by a model, it may stop being a meaningful photo. Gen Z awareness of processing: increasing awareness - Ben says even people who don’t know file formats now know phones are making decisions for them. Runner recommendation for Ray-Ban Meta: about 5K or less - V says the glasses are mostly fine for shorter, casual runs, but less ideal for longer efforts. Battery-life concern for long runs: 2.5 to 3 hours - V notes that a slower runner could exceed comfortable battery expectations on a long half-marathon effort.

Pivotal Quotes: "I think it's an image that's generated largely, majority at least, by photons that you have captured at a moment in space and time." — Sebastian DeWitt: He offers a working definition of what counts as a photo and draws the line against AI-generated fill-in imagery. "The thing that struck me is really it's such a different approach than the iPhone 16 camera, which is more about the new photographic styles... make the call on this, actually." — Allison Johnson: She contrasts Google’s memory/editing approach with Apple’s more controlled style settings and user choice. "You can just AI your way into a picture that just could be anywhere, just looks lovely." — Allison Johnson: She describes the downside of removing too much context from a photo during AI editing.

Implications: Photos are becoming editable memory artifacts, not fixed records. Expect more demand for user control, clearer editing norms, and growing tension between authenticity, convenience, and AI-generated image trust.