Colour and mood

Beginner · 5 min · Steer palette and atmosphere with dominant, support and accent colours.

  • ai images
  • colour

Colour decides how an image feels before anyone reads its details. The same room can be calm or chaotic with the same furniture — change the palette. Two or three colour decisions carry most of that weight: a dominant hue, its temperature and how saturated everything is. All three are writable.

Build a palette like an interior designer

  • Dominant — one hue owns the picture; most of the frame is its family.
  • Support — one or two related colours carry the rest.
  • Accent — a single contrasting note, used sparingly, gives the eye somewhere to land.

Name colours the way paint shops do — terracotta, sage, cream, charcoal, deep teal — rather than as “nice colours”. Concrete colour words are claims the model can render; vibes are not.

Temperature is the mood dial

  • Warm — amber, rust, gold: energy, nostalgia, comfort.
  • Cool — blue, slate, teal: calm, distance, technology.
  • Neutral — greys, beiges, off-whites: quiet and documentary.
  • Mixed — hot and cold together: tension and energy, best used on purpose.

Saturation is a volume dial

Muted palettes feel considered and quiet; vivid ones feel loud and energetic. Pastels sit at the calm end: light and low-saturation. Say which one you want — “muted and matte” is a decision, “colourful” is an invitation for chaos.

Weak prompt

a modern living room, beautiful colour scheme, very aesthetic

Stronger prompt

a modern living room in warm neutrals — cream walls, oak floor, one rust-orange cushion; muted, matte, soft afternoon light

Why the stronger one works
  • The palette names concrete colours and their relative weight: cream dominates, oak supports, rust accents.
  • “Muted and matte” fixes the saturation level; “beautiful” fixes nothing.
  • One accent colour survives the render intact and gives the composition a focal point.

A bad example

Ask: “A kitchen with perfect colour harmony and lots of colour.”

The model picks its own palette — usually a safe, generic one — and “harmony” cannot be rendered or debugged.

A better example

Ask: “A kitchen in cream and pale oak with muted sage-green cabinets, one terracotta bowl on the counter, soft morning light.”

Two families, one accent, one saturation level: a palette you can adjust by editing a single word.

Why it works

Colours are physical properties: a paint has a hue, a temperature and a finish, and generative systems render those properties literally. A named palette also holds together across the frame — the same few colours recur everywhere, which is what “harmonious” actually means. And because colour sits inside light, the two compound: warm light amplifies warm palettes. Say both when the mood has to be unmistakable.

A clinic wants images that read calm, clean and trustworthy — nothing loud. Which palette brief fits?

Practice

Choose the palette

A bakery asks for a header image for its winter menu. The brief is warm, cosy and appetising — but the last attempt came back in cool blue-grey tones that felt clinical and cold.

Which prompt gives the model the clearest direction toward a warm, appetising palette?

Choose one option
Hint

Name hues the way paint shops do — amber, cream, rust — not the feeling you hope for.

Compare the temperature: which option actually commits to warm?

Transfer

  • Brands: pick the palette once, then reuse the same three colour words in every prompt.
  • Interiors and products: muted beats vivid when you want quality cues; vivid beats muted when you want attention.
  • Mood boards: vary one dial — temperature or saturation — to explore a direction quickly.
  • Debugging: colour problems are palette problems. Restate the palette instead of regenerating blindly.

Key takeaways

  • A palette is three decisions: dominant hue, support, one accent.
  • Temperature sets emotion: warm comforts, cool calms, neutral documents.
  • Saturation is volume — muted is quiet and considered, vivid is loud on purpose.

Next

Next: Materials and texture — describing surfaces precisely enough that the render shows them. The generative AI entry covers the machinery behind these choices, and the diffusion entry explains why each run samples differently.

Finished this lesson?