A PRESSRUN-generated cyberpunk ink illustration — a courier at a desk of monitors, neon city beyond the rain-streaked window ← Back to work

AI Tool × Creative Production

Prompts in.
Press run out.

Role

Design + Build

Year

2026

Type

Self-initiated

Stack

Next.js · Replicate · nano-banana-2

Live at bulk-image-hub.vercel.app

Where it started

One prompt
at a time.

AI image tools make you babysit. Type a prompt, wait. Type the next one, wait. Fine for a single image. Murder for a hundred.

The moment you're producing a set — a campaign, a deck, a wall of variations — the tool fights you the whole way. One box, one result, one tab, repeated until your afternoon is gone.

I wanted the opposite: line the work up, walk away, come back to a finished run.

What I built

Queue a hundred.
Press once.

PRESSRUN is a bulk image studio. You stack up batches — each with its own prompt, quantity, aspect ratio and optional reference image — and run the whole stack with one click through a concurrency-managed queue.

It runs on Replicate's nano-banana-2 (Google's image model). You bring your own API key, and it never leaves your browser except through a thin proxy. No account. Prompts in, a press run out.

AI Generation Web App UX / UI Next.js Replicate Solo build

The studio

A results wall, not a feed.

bulk-image-hub.vercel.app
PRESSRUN Gallery view — control sidebar, a grid of generated cyberpunk illustrations, and an activity log

Gallery — every generation lands on the wall, with a live count badge and an activity log tracking each job.

How it runs

A production line,
one card per batch.

Two tabs. Gallery is the wall. Prompts is the line — a stack of batch cards, each a job with its own prompt, count, aspect ratio and reference image. Build the run top to bottom, then hit Generate once and the queue works through it.

PRESSRUN Prompts view — stacked batch cards, each with a prompt, aspect ratio and reference image

Prompts — the production line. Each card is a batch; the sidebar sets the press.

The part I'm proudest of

Hand the busywork
to another AI.

The best feature isn't really in the app — it's the handoff. Download a documented JSON template, give it to ChatGPT or Claude, and describe the run you want. It fills in every prompt, quantity, ratio and reference URL.

Import it back and the whole production line loads at once. Prompt-engineering becomes a supply chain — one AI briefs the other, and PRESSRUN just runs the press.

Tool, not toy

Control over every
tile in the run.

Regenerate, per image

A ↻ on every tile. Rate thumbs up or down, then regenerate every reject in one pass.

Search grounding

Toggle real-time web and image context into the model when a prompt needs the real world.

ZIP export

Pull the whole run as a ZIP with traceable filenames, backed by an activity log of every generation.

The hard part

Your key.
Your browser.
Nobody else's.

A bulk generator needs your Replicate API key — and that key is money. Asking someone to paste a paid credential into a website is asking for a lot of trust.

So the key never leaves your browser except through a thin proxy that forwards the call and forgets it. No account, no database, nothing stored.

The whole thing is a single-file vanilla-JS studio talking to a Next.js serverless proxy — small enough that you can read exactly what it does before you trust it with anything.

Bring a key.
Run a batch.

PRESSRUN is live and free to use. Bring your own Replicate key and queue your first run.

Open PRESSRUN →