PicoBerry API + MCPNow live

Guides

Image generation

Text-to-image, image-to-image, and prompt tips.

Models available

ModelBest forNotes
Gemini 3 ProHighest quality, complex compositionSlowest, most credits
Gemini 3.1 FlashFast iteration, prompt explorationGood default
Gemini 2.5 FlashBudget runs, batch experimentationLowest credit cost

Prompt structure that works

A reliable prompt covers four things in this order:

  1. Subject — what the thing is.
  2. Stylelow-poly, realistic, stylized, watercolor, etc.
  3. Composition — angle, framing, lighting.
  4. Detail — material, color palette, texture qualifiers.

Example:

a stylized treasure chest, low-poly, 3/4 view, soft rim lighting,
weathered wood and brass, muted earth tones

Image-to-image

Upload a reference image, optionally set a strength value:

  • Low (0.2–0.4) — subtle iteration, keeps composition.
  • Mid (0.5–0.7) — re-style while preserving structure.
  • High (0.8+) — treat the reference as a loose mood, generate fresh.

Common pitfalls

  • Vague subjects produce generic output. "A character" is too thin; "a stocky orc warrior, leather armor" gives the model something to work with.
  • Too many style adjectives confuse the model. Pick 2–3 strong style cues, not 8.
  • Aspect ratio is a separate control — don't try to enforce it through prompt text.