Prompting Best Practices
How to write prompts that generate better results in Omma, with worked examples for landing pages, web apps, and 3D scenes.
Why prompting matters#
Omma turns natural language into runnable output — so the clearer your description, the closer the result is to what you had in mind. A vague prompt leaves the AI guessing; a specific one lets it build exactly what you want and gives you a better starting point to iterate from.
These practices apply anywhere you prompt in Omma — the Chat app and the Canvas AI.
Core principles#
- Describe the outcome, not just the topic. Say what it's for and what it should contain, not only what it's about.
- Spell out the structure. List the sections, screens, or parts you expect (hero, pricing, footer / sidebar, dashboard, settings).
- Set the style and mood. Words like "minimal", "playful", "editorial", "retro", "corporate" steer the look; name colors and typography if you care about them.
- State your constraints. Mention responsiveness, accessibility, the framework, or performance needs up front instead of fixing them later.
- Provide real content when you have it. Real headlines, product names, and copy beat "lorem ipsum" and ground the design.
- Build incrementally. Start with the core, then add complexity in follow-up messages rather than asking for everything at once.
- Point to what to change. When iterating, reference the specific element or section ("make the hero full-screen", "change the second card's color").
A focused, well-structured prompt works better than a long, rambling one. Aim for clarity — describe the parts and the style, not every pixel.
Vague vs. specific#
A vague prompt forces the AI to make all the decisions for you:
Make a landing page for my app.
A specific prompt gets you something usable on the first try:
Build a landing page for "Focusly", a focus-timer app. Include a hero
with a bold headline and an email signup, a three-column feature grid
with icons, a short testimonials row, and a footer. Use a calm, modern
look — off-white background, one deep-teal accent, generous spacing,
and a clean sans-serif.
Examples by what you're building#
Landing pages#
Name the brand, list the sections in order, and set the visual tone.
Create a landing page for "Northwind", a B2B analytics platform.
Sections, in order: sticky nav with logo and a "Book a demo" button;
hero with a headline, subheadline, and two CTAs; a logo cloud of
customers; a bento-style feature grid; a pricing section with three
tiers; an FAQ; and a footer. Professional and trustworthy — dark theme,
indigo accent, subtle gradients, rounded cards. Make it responsive.
Web apps#
Describe the pages/screens, the layout, and the key interactions. Omma builds multi-file apps, so you can ask for several views.
Build a multi-page task manager app. Include a sidebar nav (Today,
Upcoming, Projects, Settings), a main board with draggable task cards
in three columns (To do, Doing, Done), a "new task" modal with title,
due date, and priority, and a settings page to toggle dark mode. Persist
tasks in local storage. Keep it keyboard-accessible and responsive.
3D scenes#
Call out geometry, materials, lighting, camera, and any animation or interaction.
Create a 3D scene of a small floating island with a low-poly tree, a
rock, and a tiny house. Soft pastel materials, warm directional light
with soft shadows, and a subtle ambient fill. Slowly rotate the island,
and let me orbit the camera with the mouse. Add a faint gradient sky.
The prompts above generate runnable projects. For standalone images, video, 3D models, materials, or audio, see the Studio pages — each modality has its own prompting notes.
Iterating on a result#
You rarely get everything in one shot — that's expected. Refine in conversation, one change at a time, and be specific about which part:
Make the hero full-screen with a background image, tighten the spacing
between feature cards, and change the accent color to amber.
The pricing cards are too tall — reduce the padding and move the
"Most popular" badge to the top-right corner of the middle card.
Small, targeted instructions are easier for the AI to apply cleanly than a long list of unrelated changes.
Going further#
- Lock in a look with a design system. Provide a
design.md(or create one in Omma) so every generation follows your palette, typography, and spacing. See Designing with AI. - Reuse guidance with custom skills. Save recurring instructions — a design reviewer, an accessibility checker — and invoke them with
/your-skill. See Custom AI Skills. - Bring real assets. Attach images, video, or 3D models so the AI embeds your actual content instead of placeholders.
- Generate many options at once. On the Canvas, ask for several variations in one prompt to explore directions in parallel before committing to one.
Before you hit send, ask yourself: did I name the purpose, the structure, the style, and any constraints? If yes, you'll get a much stronger first result.