Writing Effective Prompts

A great prompt sounds like clear direction you’d give a teammate. Keep it simple, specific, and conversational.

Start with the essentials

  • Say what you want – “Animate the Logo layer scale from 0% to 100% over 12 frames with ease-in.”
  • Name the target – mention the composition, layer, or property you have in mind.
  • Give numbers when they matter – frame counts, positions, colors, durations.

Paint the picture

  • Add the “why” when it helps: “Create a camera rig centred on Controller so I can orbit around the hero card.”
  • Reference existing elements: “Match the font and size already on Title_Main.” Atom reads the project and follows the breadcrumbs.

Provide visual context

Attach files to give Atom visual reference when words alone aren’t enough:

  • Data visualization – Attach CSV or Excel files and ask Atom to generate animated bar charts, line graphs, or data-driven motion graphics
  • Design references – Drag in a mood board or client comp and ask Atom to match colors, spacing, or animation style
  • Technical specs – Attach a PDF brief with exact dimensions, timing charts, or font specifications
  • Audio sync – Include an audio file and ask Atom to analyze beats or create markers for sync points
  • Asset prep – Send texture maps, logos, or sprite sheets for Atom to import and integrate

Example prompts with data:

Attach sales_data.csv and create an animated bar chart showing quarterly growth
Use this CSV to drive the height of these shape layers over time

Files you attach are sent through OpenRouter to the AI model. Project footage stays local unless you explicitly attach it.

Build momentum

  • Break bigger goals into short prompts and confirm each step.
  • Stay in the same conversation when you’re iterating. Atom remembers what just happened.
  • Need a different approach? Ask for options (“Suggest three variations on this animation”).

Try voice prompting

LLMs thrive on details, and it’s much easier to provide context by speaking than typing. Voice input lets you naturally describe what’s in your head without condensing your thoughts. Give handy.computer a try. It’s a free, open-source transcriber that runs locally on your machine.

When results miss the mark

  1. Add more detail to your prompt.
  2. Tighten the scope (“only the selected layers”, “just this comp”).
  3. Rephrase with the outcome up front.

Quick reminders

  • Use clear nouns, numbers, and intent.
  • Reference actual layer names or patterns Atom can search for.
  • Skip vague phrases (“fix it”, “make it better”).