/3d-illustrations

One polished 3D object per image, a soft studio light, rounded bevels, and a tiny friendly face. This makes a product feel finished and premium. The object style came from xoox.fun (chrome coins, fanned cards, a coin stack on a ring tray). Never copy their silver, their pill-eye mascot or their logo. Every product gets its own colour, material and face.

1. Pick the product's kit (write it down before any prompt)

Save these 5 lines as docs/ART-3D.md in the repo:

• Material: one family per product, e.g. glossy candy enamel, anodized aluminium, frosted glass with an inner glow, matte ceramic with a gloss rim, soft-touch rubber, polished brass. Silver/chrome only if the user asks for it.
• Palette: the product's own brand colours as hex. One main colour for the objects, one accent for details, one rim-light colour. Check other products' ART-3D.md files so no two products in one event share a palette.
• Face: the product's own tiny mark, e.g. two curved ink drops, a small smile line, one round eye, a visor slit. Keep it on objects where a face fits, not every one.
• Background: transparent for UI objects (they sit on light or dark surfaces), or a seamless studio sweep in a tint of the palette for hero shots.
• Subjects: the product's own metaphors (for comics: a comic issue, a speech bubble, a pen nib; for a game: a crown, a card, a tower, a chest; for DeFi: a vault, a coin stack, a key).

2. Prompt template (reuse the same style line in every prompt, so the set matches)

Create an image: <subject, one object, simple composition>. <MATERIAL> 3D render in <main hex> with <accent hex>
details and a soft <rim hex> rim light, soft studio lighting, crisp reflections, rounded bevelled edges, gentle
toy-like proportions, <face line>, centred, transparent background (no background, no ground shadow), square
1024x1024, no text, no letters, no logos.

Ask for 8–12 subjects: one hero, 3–5 feature objects, empty-state, success/error state, a badge or trophy, and a logo mark candidate.

3. Generate (ChatGPT in the user's real Chrome)

Use the /chatgpt-assets workflow: load the Claude-in-Chrome tools in ONE ToolSearch call, open chatgpt.com in a new tab (the user is signed in), use a fresh chat per product, send one image per message, and download each image as a blob into ~/Downloads named <product>-3d-<subject>.png. Don't upload reference images (another brand's art). If ChatGPT refuses or rate-limits, stop and say so.

4. Check the set before wiring it in

• Make a contact sheet of all images and look at it. Regenerate any image with the wrong colour, extra objects, text, a cut-off subject, a cast shadow on a transparent image, or a face that doesn't match the kit.
• Check each object at 48 px. It must still read as the subject.
• Convert to webp (transparent: cwebp -q 85 -alpha_q 90 -resize 640 0; hero: -q 82 -resize 1400 0), keep files under ~400 KB, and put them in the app's public art folder.

5. Wire into the UI

• Objects go in framed feature cards, hero corners, empty states, badges and toasts. One object per card, generous padding, no glow halos and no sticker borders.
• Gentle float (2–6 px over 4–7 s) or a slight parallax, turned off under prefers-reduced-motion. Width and height set, so nothing shifts as the page loads; alt text describes the object.
• On dark UIs, add a soft radial light behind the object (one tint of the palette) so its rim reads.
• README credit: "3D art generated with ChatGPT".

Related: /chatgpt-assets (generation mechanics), /dribbble-frontend (layout and finish).