There are three ways to create your 3D pop-up — text prompt, photo upload, or curated presets. Here's how to get the best result from each.
🎨 Presets — choose from our curated library. Loads instantly in the browser. Best for a reliable, high-quality result in seconds.
📸 Photo to 3D — upload a photo of a real object and our AI reconstructs it in 3D. Best for personal items and keepsakes.
✏️ Text prompt — describe anything in words. Most creative freedom. Takes 1–3 minutes to generate.
Our AI reconstructs a 3D mesh from a single photo. The quality depends heavily on the source image — here's what works.
File requirements: JPEG, PNG, or WEBP · Max 10 MB · Min 512 × 512 px recommended for detail
Generation takes 1–4 minutes depending on image complexity. The resulting 3D model is saved permanently to your card — it won't need regenerating.
Model format: GLB (binary glTF 2.0) — the industry standard for real-time 3D on the web and in AR.
File size: Preset models are 3–8 MB · Custom text-to-3D is 3–6 MB (preview) · Photo-to-3D is typically 5–15 MB
AR viewer: Works in any modern mobile browser (iOS Safari 15+, Chrome on Android). No app download required. The AR experience uses WebXR / device camera — users must grant camera permission when prompted.
Loading time in AR: 3–8 MB models load in ~3–8 seconds on a standard 4G connection. Larger photo-to-3D models may take 10–20 seconds on slower connections.
Models are served from our CDN and cached after the first load. Repeated opens are near-instant.
A great prompt usually follows this pattern:
Example: "A blooming red rose bouquet, photorealistic, rich colours"
Example: "A golden birthday cake with lit candles, cartoon 3D style, warm and festive"
Example: "A small Christmas tree with ornaments and snow, cute low-poly style, green and gold"
Keep it to one main object. Think of it as describing a figurine or a trophy someone might sit on a shelf.
3D AI models are best at standalone objects — things you could pick up and hold. Single, clear subjects with recognisable forms generate the most impressive results.
Adding a style description after your object helps the AI choose the right look:
These either produce poor 3D shapes, or may be filtered for safety/copyright reasons. Cultural symbols (e.g. a koru, tiki, or regional emblem) often don't have a universally understood 3D form — the AI produces something generic or distorted. If your first attempt doesn't look great, try simplifying the prompt or switching to a style keyword like "cartoon 3D".
Even with a well-written prompt, AI-generated 3D models can occasionally produce unexpected results — slightly melted shapes, unusual proportions, extra limbs, or details that weren't in the prompt. This is a normal characteristic of generative AI, not a bug.
Some people enjoy the quirky, abstract character of AI outputs. If you'd prefer a guaranteed result, try one of the curated preset models — each one is quality-checked by the POP! team.
If the first result isn't what you expected, you can retry from your card page. Simpler prompts — one clear object, one style keyword — tend to produce the most consistent results.
Need a starting point? Each of these follows the prompt formula and works reliably: