I made artifacts/image.png: a 512 × 512 PNG of a tangerine ghost bust drawn as 32 × 32 pixel art and scaled up 16× with hard edges. I looked at the rendered image, and decoding the file confirmed the size, 8-bit RGB format and 10 colours. ffprobe isn't installed here, so I checked the file directly instead.

What's in it:

• One ghost, front-facing and centered, with big glossy black eyes, pink blush and a small smile.
• Headphones with a grey band and red-orange ear cups with gold detail.
• A blue crew-neck tee whose shoulders reach the bottom and side edges, with a tiny white "UI" logo on the chest.
• Thick black outlines, 1 grid pixel (16 px) wide, around every shape. The background is solid #0c1610 with four tiny green specks. There are no gradients, blur, text, frame or scenery.

I didn't use an image-generation model. I placed every pixel in a short Python script (tools/draw_iu0460.py, no extra packages needed), so the grid and colours match the brief exactly. Running python3 tools/draw_iu0460.py artifacts/image.png regenerates the image.

Requirements not fully met (also recorded in README.md):

• The outlines are pure black (#000000), which isn't in the 14-colour list. The brief asks for black outlines, so it's the only colour I added.
• I haven't compared it with the rest of the 787-piece set, so whether it fits that set's look is your call.
• Because it's a bust, the shirt hides the ghost's wavy hem. It reads as a ghost from its rounded, flared body and big eyes.
• The "UI" logo is 5 × 3 grid pixels, the smallest size at which it stays readable.

The script and README are committed. The image is not committed: it's left untracked for the upload step, and a gitignore rule outside this repository happens to mark the artifacts/ folder as ignored.