CHARACTER: Canvas 1024x1024, transparent background, PNG.

Style: 1930s rubber-hose cartoon in the Cuphead style (thick wobbly black ink lines, flat muted retro colours, soft watercolour shading, light film grain), following the creator's reference image as closely as possible. Draw ONE friendly, non-human, object-headed cartoon character in the Cuphead tradition (for example a mug or teapot-style head with a simple rounded rubber-hose torso, NOT a human body: no human muscles, collarbones or anatomy, smooth bean-like shoulders).

Pose: facing straight at the viewer, head centred on x=512, both pie-cut eyes level at y=425 and looking at the viewer.

Geometry: crown (top of the head dome, the rim of the object head) at y=210; head 350px wide (x=337 to x=687); chin or bottom of the head at y=665; a short narrow neck with its narrowest point at y=700, about 120px wide (x=452 to x=572); the neck meets the shoulders at y=740; rounded shoulders and upper chest run off the bottom edge. The top of the head must be a clean, open dome with nothing sticking up, so a big hat can sit on the crown.

Arms down and out of view, nothing in front of the chest.

Leave out: mouth (no mouth line, leave a smooth blank area around y=560 centred on x=512), all clothing (bare neck and shoulders, no collar, no shirt, no bow tie), chains and necklaces, hats. A simple nose is allowed.

Output: character_base.png.

Weights file: character_weights.json with {"character_base.png": 1}.

BACKGROUND: Canvas 1024x1024, opaque PNG, full bleed.

Style: 1930s rubber-hose cartoon in the Cuphead style, hand-painted watercolour backgrounds, muted retro palette, film grain and soft vignette.

Each background is a different classic platformer game level drawn as a flat-perspective side-scrolling scene: e.g. grassy hills with floating platforms, a carnival boardwalk, a haunted forest, a candy land, a cloud kingdom, a river with log platforms, a mine, a pirate dock, a snowy mountain, a casino hall, a factory with gears, a desert with ruins.

Keep the centre area (x=300 to x=724, y=180 to y=1024) low in detail and contrast, as the character sits there; put platforms, coins and props near the edges and the horizon. No characters, no text, no UI or score counters.

Outputs: background_01.png to background_12.png.

Weights file: background_weights.json mapping each file name to 1. Also write artifacts/base_b.png and artifacts/base_c.png: two more takes of the SAME character on the SAME template and pose, each a SEPARATE, FRESHLY GENERATED image (run the image tool again for each one; never copy, edit, recolour or re-save base.png), so that if one comes out too big, too small or off-centre another may fit.

Write artifacts/base.png, artifacts/background_01.png .. background_12.png, and artifacts/traits.json: {"backgrounds":[{"file","name","weight"}...]} with weights summing to 100.