January 28, 2026
Pretty pics, spicier comments
Trying to craft AI images that are worth displaying to end users
Travel app picks pretty pics; commenters cry copyright traps and designer doom
TLDR: A travel app parses trip queries into places and shows curated photos instead of AI images. Commenters brawled over “just use Wikipedia,” warned that Unsplash’s fine print bites at scale, and dropped doom posts about designer jobs—raising real questions about copyright, curation, and who gets paid for pretty pictures.
An AI travel app, Stardrift, wants your trip homepage to look like a postcard: when you type “sfo-jfk,” it uses a large language model (an AI that reads text) to figure out you mean New York, then shows a hand‑curated photo by real photographers—not AI slop. The founder ditched auto‑generated images (too ugly, too pricey) and random Google grabs (copyright minefield) for a crafted “place → picture” system that feels intentional.
Cue the comments section meltdown. One camp says, “Why not just use Wikipedia?” with a straight‑up link to Wikipedia’s Deadvlei. Another camp yells “Licensing nightmare!” warning that Unsplash looks free until you scale, then the terms bite. A third wave brings the apocalypse: “This will kill graphic designers.” Meanwhile, a spicy post gets [flagged], and a moderator pops in with a sheepish “oops, wrong title” mea culpa—classic thread energy. People joked the easiest algorithm is copy‑paste from Wikipedia, roasted “vibes over rights,” and debated whether tasteful curation is the antidote to AI dreck or just a slick shortcut around paying creatives. TL;DR: pretty pictures, messy feelings, and a reminder that in 2026, the hardest part of shipping is wrangling the comments
Key Points
- •The author aims to convert freeform travel queries into curated destination photos for an AI travel app (Stardrift).
- •They avoid AI-generated images and generic web searches due to quality, cost, copyright, and risk.
- •They model outputs as a list of ‘places,’ each with a name and type (city/region/country).
- •An LLM (e.g., Haiku) parses inputs like “SFO-JFK” into structured places, handling multiple destinations.
- •They begin building a place-to-photo mapping database, seeded by real app queries and human-curated images.