Most bad AI images aren't the model's fault — they're the prompt's. “A beautiful product photo” could mean a hundred things, so the AI picks one at random and you pay for the lottery ticket. Prompt engineering sounds technical, but it's just structured describing: give the model the five decisions it can't make for you, and it stops guessing. At ₹15 per finished image, every avoided re-roll is money kept.
The five parts, with an example
Here's a complete prompt built with the formula — a D2C candle brand's hero shot:
- SUBJECT — the exact thing: “hand-poured soy candle in an amber glass jar, cream label with minimal black text”.
- STYLE — the visual language: “warm lifestyle product photography, premium D2C brand aesthetic”.
- SCENE — where and what's around: “on a light oak table beside dried eucalyptus, soft linen cloth underneath”.
- LIGHT — time, source, mood: “soft morning window light from the left, gentle shadows, cozy glow”.
- CAMERA — angle and framing: “straight-on hero shot, centered, shallow depth of field, 4:5 portrait crop”.
Common mistakes, fixed
| Mistake | Why it fails | Fix |
|---|---|---|
| “Make it premium” | Vague — premium means different things | Name the cues: “matte black, gold foil accents, dark background” |
| Contradictory adjectives | “Cozy yet dramatic” pulls two ways | Pick one mood per generation |
| Forgetting the background | AI fills the void randomly | Always include the scene part |
| No negative guidance | Unwanted elements appear | Add “no hands, no watermark, no extra objects” |
| Changing everything on retry | You learn nothing | Change one part at a time |
Indian-context prompting: be specific, get specificity
Generic prompts produce generic imagery — often Western-default. “A family celebrating” tends toward imagery that doesn't look like your audience. Name the specifics: “a North Indian family celebrating Diwali, marigold torans on the door, diyas on the balcony railing, evening”. Festivals, clothing, architecture, street scenes — the model knows them, but only if you ask. This is the cheapest quality upgrade in AI imagery: ten extra words of cultural specificity.
The same applies to products. “Ayurvedic face oil” gets you a prettier result than “face oil” because it anchors the styling — brass bowls, neem leaves, warm wood. Your product's context is a prompt asset; spend it.
Iterating without burning money
Treat generation like a conversation, not a slot machine. First order: your best five-part prompt. Review against the brief part by part — subject right? light wrong? Then revise only the failing part. The ₹49 4-pack is the iteration tool: four variants of one concept in a single order, so you can test “morning light vs evening glow vs studio softbox vs golden hour” side by side instead of guessing sequentially. Check the pricing page for the full menu.
Steal this template
Copy, fill, generate: “[SUBJECT with exact details], [STYLE / visual language], [SCENE with surroundings], [LIGHT with time and mood], [CAMERA angle and crop]. No [things to avoid].” Keep a running doc of prompts that worked — your best prompts are reusable assets. A seller with forty SKUs and one proven prompt template has a catalog; a seller re-rolling from scratch every time has a hobby. Review your template quarterly: as your brand evolves, the style and light parts are usually what need updating, while subject and camera stay constant.
Test the formula for ₹15
Write your five parts, order a single image, and see how close the first try lands.
Try a five-part prompt