The news, 365 days behind — on purpose Delayed live · replaying 2025

One Year Ago.AI

Remember how fast this is.

22JAN2023replayed
one year on
open weightsbrycedrennan

Show HN: New AI edits images based on text instructions

A Python library wrapping the InstructPix2Pix model lets developers alter photos by typing commands like 'make it winter' or 'remove the cars'.

A new open-source Python library called imaginAIry, wrapping the InstructPix2Pix model, allows developers to edit images by typing plain-text instructions.

The tool, posted to Hacker News by developer Bryce Drennan, lets users run commands like ‘make it winter’ or ‘remove the cars’ to alter photographs. Drennan demonstrated edits turning a portrait of the Girl with a Pearl Earring into a cyborg or making the Golden Gate Bridge snowy. The library builds on Stable Diffusion and works on GPU-equipped systems or Apple M1 Max hardware, with edits taking about two seconds on a 2080 Ti.

The HN thread sparked discussion about the model’s limitations. One commenter noted that AI tools like this are good at replacing textures but often misplace objects, such as adding tiny fireworks to portraits or giant trees under bridges. Another user praised the library’s accessibility, saying they ‘wrote Bryce a thank-you email in December’ after using it for a personal project. Drennan also shared a Google Colab notebook for those lacking local GPU resources.

B
bryced@bryced

says 'this works suprisingly well' and provides examples of transformations like making the Golden Gate bridge snowy or turning a portrait into a cyborg

S
sandworm101sandworm101

observes that AI is good at replacing textures but struggles with object insertion, citing tiny fireworks and monster trees as common failures

P
PaulMestPaulMest

says he sent a thank-you email to the developer after using imaginAIry to make an advent calendar, calling it the best-fitting SD framework

One year later — open only if you can handle spoilers

ImaginAIry quietly continued as a niche Python library while the broader field of instruction-based image editing exploded with diffusion models like Stable Diffusion XL and DALL-E 3. The tool never saw wide commercial adoption, but its open-source code influenced other editing pipelines.

Replay thisPost on XRedditHNLinkedIn

The Weekly Replay · free by email

This week, one year ago — every Sunday.

One email each Sunday: the week's replayed AI news, with the one-year-later annotations included. Written like it's breaking — dated like it isn't.

Free · double opt-in · unsubscribe anytime · privacy