version: "1.0.0" name: image-prompting description: Write stronger prompts for Nano Banana 2, Nano Banana Pro, and other image models using a structure distilled from Google's Nano Banana prompting guidance.

Image prompting

Use this skill before calling mcp-image when the request is vague, the composition is under-specified, or the user wants a more intentional result.

This skill does not generate images. It helps the agent turn a rough request into a cleaner prompt without bloating the MCP call itself.

The prompt structure here is distilled from Google's Nano Banana guidance and adapted for day to day agent use.

Core frame

Build the prompt around three pieces:

1. Subject

• What is the focal point?
• What materials, colors, textures, scale, or defining traits matter?
• If a person or recurring character is involved, what makes them recognizable?

2. Context

• Where is this happening?
• What belongs in the foreground, midground, and background?
• What time of day, atmosphere, weather, or mood should show up visually?

3. Render intent

• Should it feel photographic, cinematic, graphic, product-focused, editorial, illustrative, or tactile?
• What lighting shapes the scene?
• If the request is photographic, what camera angle, lens feel, depth of field, or crop matters?

Rules

• Preserve the user's intent. Add detail only where the user left ambiguity.
• Prefer positive instructions over long exclusion lists.
• Write one clean visual paragraph, not a bag of comma-separated tags.
• If the user already gave a precise prompt, keep it tight.
• If text must appear inside the image, put the exact text in quotes.

High-value upgrades

Lighting

Turn soft words into visual direction:

• "soft morning light through tall windows"
• "hard noon sunlight with crisp shadows"
• "warm golden-hour rim light"
• "cool overcast daylight with low contrast"

Composition

Give the model spatial clarity:

• "subject centered with generous negative space on the right"
• "close product framing with shallow depth of field"
• "wide cinematic scene with layered foreground and distant haze"
• "low-angle medium shot"

Materials and texture

Name the surface:

• "brushed aluminum"
• "matte paper stock"
• "weathered terracotta tile"
• "polished glass with soft reflections"

Editing existing images

When an input image is provided:

• say what to preserve
• say what to change
• describe the change relative to the source image

Useful phrases:

• "Keep the existing composition and lighting."
• "Preserve the original palette and materials."
• "Only change the background to..."

Character consistency

When the same character needs to stay recognizable:

• include at least three stable visual markers
• repeat concrete identifiers

Example: "same character with a shaved side haircut, round tortoiseshell glasses, and a mustard utility jacket"

Purpose-specific emphasis

Product image

• clean setting
• clear silhouette
• controlled reflections
• commercial polish

Poster or cover

• strong focal hierarchy
• room for typography
• bolder mood

Social image

• crop-friendly framing
• fast readability
• stronger contrast

UI mockup or diagram

• flatter composition
• cleaner shapes
• reduced visual noise

Output style

Default output:

• one polished prompt paragraph
• optionally one short note with recommended aspect ratio or quality preset

Example

Input: "banana on a table"

Prompt: "A single ripe banana resting on a clean matte white tabletop, photographed in a quiet studio setting with soft daylight from the left and a gentle shadow falling to the right. Tight product-style framing with generous negative space, subtle surface texture, and crisp peel detail, rendered as a minimalist commercial photograph with natural color, restrained contrast, and a calm editorial finish."