Sunday, June 7, 10:33 AM. A Stripe email landed in my inbox — far from my first online payment, yet the most satisfying one in years: “Congratulations, Bernard Huang! You’ve just received your first payment.” Five dollars. A stranger uploaded a portrait, my pipeline converted it into an animated pixel character, and Stripe cleared the checkout while I was away from my desk. PixelForge is live.

TL;DR

PixelForge turns any portrait photo into a game-ready pixel sprite pack — a strict 4×4 walk sheet, individual transparent PNG frames, directional strips, and looping walk GIFs — for $5. No account, no subscription, automatic refund if a pack fails QA.

  • Generated by AI, finished by code: The image model draws the 4×4 sheet; deterministic Python removes the key background, aligns all 16 frames to a shared foot baseline, and slices engine-ready assets.
  • Built in days using the plan-3x-build-once workflow and GPT 5.5.
  • Shipped under Ship or Die rules: Launch a paid product in 30 days or get permanently booted.
  • First live customer on June 7: $5 Stripe receipt below, plus a playable demo game (Stonk Runner) built directly from the output packs.

What five dollars buys

One portrait goes in; a complete sprite bundle comes out: a 4×4 sprite sheet (down, left, right, up), 16 individual transparent PNG frames, directional strips, and looping walk GIFs — formatted to drop straight into Godot, Unity, Phaser, GameMaker, or any web canvas, complete with engine import notes. There is no account wall and no recurring subscription; checkout is a single Stripe button, backed by a money-back guarantee if the pack misses QA. The pitch is simple: put anyone into your game — yourself, your co-founder, or the friend who told you your indie game idea was dumb.

How it works: generated by AI, finished by code

PixelForge pipeline: a source portrait becomes a strict 4×4 sprite sheet, then deterministic Python repairs and extracts a transparent walking GIF.
The production line: source photo → strict 4×4 sheet → repaired, transparent, walking. (Yes, that's me.)
  1. Upload a portrait. Any clear front-facing photo works.
  2. AI generates the sheet. The model renders a 4×4 directional walk grid in a consistent pixel-art style.
  3. Deterministic Python finishes the pack. Code strips the key color, snaps every frame’s feet to a uniform baseline, and exports clean transparent assets.

Step 3 is the actual product. Raw AI sprite sheets are 90% impressive and 100% unusable in a real game engine: background artifacts bleed into edges, characters bob up and down off their baseline as they walk, and cell boundaries drift by a few pixels. Almost every “AI sprite generator” demo stops at step 2 and hands you a flat PNG you still have to clean up in Aseprite. The unglamorous Python post-processing — the part a model cannot vibe its way through — is what makes the output immediately playable.

Stonk Runner: the playable proof

Claiming an asset pack is “game-ready” is cheap, so I built a game to prove the files work untouched. In one evening of coding, I ran single portraits of Jensen Huang, Elon Musk, Warren Buffett, and Leopold Aschenbrenner through PixelForge and dropped the raw zip outputs — zero manual pixel edits — straight into Stonk Runner, a 60-second arcade autorunner across NVDA’s May 25, 2023 earnings gap-up.

Stonk Runner's animated select screen: nine real markets to pick from, the four CEOs — Jensen, Elon, Warren, Leopold — with free grab-pack buttons, and the Ring the Opening Bell start button.
The select screen, live: pick a market, pick a CEO, grab their pack free, ring the opening bell. Every character started as one photo.

Play Stonk Runner for free — and grab any of the four CEO sprite packs directly from the character select screen at no cost. Drop Jensen’s pack into Godot, Unity, or Phaser tonight to test the exact file structure before spending a dime, then forge your own face for $5.

The launch trailer is my favorite artifact of the build because I didn’t edit a single frame. After generating the soundtrack (“Rocket Twenty-Six”), I handed the MP3 to GPT 5.5 with a one-line prompt: make me a music video for this. It cut all 146 seconds below programmatically to the 152 BPM beat.

The trailer — 100% AI-generated by GPT 5.5 from nothing but the mp3.
“Rocket Twenty-Six” — the track that became the brief. Margin calls have never slapped harder.

PixelForge is also live on Product Hunt today (currently sitting at #14). If the pipeline or the game made you smile, an upvote there helps indie game devs find it.

The build: days, not weeks

PixelForge went from blank repo to live checkout in a few days using my standard plan-3x-build-once loop: three independent model plans reconciled into one spec before GPT 5.5 wrote the code. Prompting the image generator took hours; hardening the deterministic Python slicer took days — reliably detecting grid lines, keying out backgrounds without chewing into dark outlines, and locking 16 frames of feet to a single floor line. The web wrapper stayed intentionally minimal: one landing page, one file upload, one Stripe checkout button, and a live public bench showing recent generations.

Ship or Die

Deadlines shipped this product as much as code did. I joined Ship or Die — Marc Lou and Jack Friks’ accountability community built around a single rule: launch a startup in 30 days or get permanently banned. Entry costs $269, progress check-ins are daily, and missing your deadline gets you thrown overboard in public with no refund.

The irony isn’t lost on me given my own benchmark on stakes priming: threatening an LLM doesn’t improve its output, yet paying $269 for a Discord server to threaten me got a working product out the door in under a week. Forcing the scope down to “one core feature and a live Stripe button” is the exact antidote to builder perfectionism.

The first $5

Stripe email: Congratulations Bernard Huang, you've just received your first payment — $5.00, June 7, 2026.
The first $5, June 7 at 10:33 AM. Framed, obviously.

That first $5 came from a fellow Ship or Die builder testing the pipeline. Five dollars doesn’t move a P&L, but it changes the ontological status of the project. Before 10:33 AM on June 7, PixelForge was a script that worked on my own face. After 10:33 AM, it was a commercial product that took a stranger’s photo, delivered a working game asset pack, and cleared payment with zero human intervention — a welcome contrast to PalmAura, which died in App Store review without ever reaching a single user.

Try it at pixel-forge.net. Upload a face, grab a sprite pack for $5, and I’ll follow up with full traffic and unit-economics receipts once the launch dust settles.