How to Write Nano Banana JSON Prompts

Turn a general image idea into a structured first draft.

{
  "task": "generate_image",
  "subject": "[original subject or object]",
  "goal": "create a polished visual draft for [channel]",
  "scene": "[setting, background, and context]",
  "composition": "[framing, crop, focal point]",
  "style": "[broad visual direction, not a living artist or protected brand style]",
  "lighting": "[time of day, contrast, shadow softness]",
  "constraints": ["no NSFW", "no protected logo copy", "no fake affiliation", "no public figure likeness"],
  "output_check": ["subject is clear", "scene matches the task", "no misleading claims"]
}

It forces the prompt to name the job, visual choices, and safety boundaries before generation.

Create ecommerce-safe product imagery for owned or authorized products.

{
  "task": "product_photo",
  "subject": "[owned or authorized product type]",
  "preserve": ["shape", "material", "label area", "key product features"],
  "scene": "clean [studio/lifestyle] background for [store or ad channel]",
  "camera": "three-quarter angle, realistic scale, product centered",
  "lighting": "soft commercial studio light with natural shadow",
  "constraints": ["no fake marketplace badges", "no third-party logos", "no unsupported performance claims", "do not alter product features"],
  "output_check": ["product remains accurate", "background supports the channel", "claims are not invented"]
}

It protects product truth while still giving the model enough direction for a usable product-photo draft.

Edit an uploaded or referenced image while preserving approved details.

{
  "task": "image_to_image_edit",
  "reference_use": "use my authorized image as the visual reference",
  "keep": ["main subject identity if fictional or authorized", "pose", "key colors", "important object details"],
  "change": ["background to [new setting]", "lighting to [direction]", "crop to [format]"],
  "do_not_change": ["body shape", "private identity", "product facts", "logos or labels unless provided and authorized"],
  "constraints": ["no adult edits", "no deceptive before-after proof", "no public figure imitation", "no fake endorsement"],
  "output_check": ["kept fields stayed stable", "changes are limited", "no private or misleading transformation"]
}

The keep/change split makes the edit brief auditable and reduces accidental identity, product, or claim drift.

Use references for visual control without promising pixel-perfect consistency.

{
  "task": "reference_guided_image",
  "reference_rights": "owned, licensed, client-approved, or fictional reference",
  "stable_anchors": ["silhouette", "color palette", "material texture", "key accessory", "overall mood"],
  "allowed_changes": ["background", "lighting", "camera angle", "layout format"],
  "scene": "[new scene] while preserving the stable anchors",
  "constraints": ["no private person imitation", "no celebrity likeness", "no protected trade dress copying", "no consistency guarantee"],
  "output_check": ["anchors are recognizable", "changes are allowed", "result is not presented as proof"]
}

It frames reference use as controlled guidance, not identity cloning or official model behavior.

Draft layout-aware poster visuals before adding verified copy.

{
  "task": "poster_concept",
  "event_or_offer": "[verified event, launch, or campaign topic]",
  "visual_theme": "[original theme and mood]",
  "layout": {"format": "[square/portrait/landscape]", "text_zone": "leave [area] clean for approved copy"},
  "elements": ["original shapes", "non-branded props", "safe background motif"],
  "constraints": ["no fake sponsor logos", "no false organizer approval", "no ticketing proof", "no mature or unsafe content"],
  "output_check": ["copy area is usable", "claims are verified separately", "no invented badges"]
}

It separates visual ideation from factual copy, sponsors, and approval claims.

Create original character or mascot prompts safely.

{
  "task": "character_concept",
  "character_type": "original [mascot/avatar/game NPC]",
  "personality": "[friendly, calm, bold, curious]",
  "visual_traits": ["[shape language]", "[color palette]", "[outfit or prop]"],
  "pose_scene": "[pose] in [fictional setting]",
  "constraints": ["fictional character only", "no celebrity or private person likeness", "no protected IP copy", "no adult/NSFW character framing"],
  "output_check": ["character is original", "traits match the brief", "no recognizable protected character"]
}

It keeps the character brief original and reusable without leaning on protected IP or identity imitation.

Structure high-detail jewelry prompts without fake appraisal or authenticity claims.

{
  "task": "jewelry_product_image",
  "item": "[ring/necklace/earrings/bracelet]",
  "material_finish": "[approved metal color, finish, texture]",
  "stone_detail": "[stone color, cut, setting] if known; otherwise describe visual color only",
  "scene": "macro product photo on [surface] with controlled reflections",
  "camera_lighting": "close-up lens, crisp highlights, soft shadows",
  "constraints": ["no counterfeit luxury styling", "no fake certificate", "no carat or purity claim unless verified", "no official designer affiliation"],
  "output_check": ["visual draft only", "materials are not overclaimed", "no fake authenticity proof"]
}

It captures jewelry detail while keeping the output inside visual drafting boundaries.

Generate campaign-safe visual concepts for owned offers.

{
  "task": "social_ad_visual",
  "offer": "[verified product, service, or creator offer]",
  "audience": "[target viewer]",
  "scene": "[original visual metaphor or lifestyle context]",
  "format": "[platform ratio] with room for approved headline",
  "style": "clean modern ad visual, no protected brand imitation",
  "constraints": ["no fake testimonials", "no false scarcity", "no unverified results", "no misleading platform badges"],
  "output_check": ["offer remains truthful", "space for copy exists", "no deceptive proof elements"]
}

It keeps ad visuals separate from claims that need human review and proof.

Create safe YouTube or article thumbnail drafts with controlled focal points.

{
  "task": "thumbnail_visual",
  "topic": "[video or article topic]",
  "main_visual": "[original object, scene, or fictional character]",
  "emotion": "[clear but not deceptive expression or mood]",
  "layout": "bold focal point, clean negative space for title overlay",
  "color": "[high contrast palette]",
  "constraints": ["no public figure likeness", "no fake shock event", "no medical or financial proof claims", "no protected character copy"],
  "output_check": ["topic is readable", "title zone is clear", "visual is not misleading"]
}

It creates an attention-focused brief while avoiding deceptive or identity-based thumbnail tactics.

Replace or clean a background without changing the subject facts.

{
  "task": "background_replacement",
  "reference_use": "authorized subject image",
  "keep": ["subject shape", "scale", "key color", "visible details"],
  "replace_background_with": "[new safe background]",
  "lighting_match": "match shadows and reflections to the new scene",
  "constraints": ["do not alter product facts", "do not add third-party logos", "do not create fake location proof", "no adult or unsafe setting"],
  "output_check": ["subject facts unchanged", "background is plausible", "no false context is implied"]
}

It controls the most common edit request without allowing the background to create false context.

Explore original logo directions without copying protected marks.

{
  "task": "logo_concept",
  "brand_use": "original concept exploration for [project/category]",
  "symbol_idea": "[abstract shape or metaphor]",
  "typography_direction": "[broad type personality], not a copied brand wordmark",
  "palette": "[approved colors]",
  "constraints": ["no existing brand clone", "no certification badge", "no fake partnership", "not legal trademark clearance"],
  "output_check": ["mark is original", "no protected logo resemblance", "ready for human designer review"]
}

It treats generated marks as ideation, not trademark-ready assets or official badges.

Improve a weak generation by editing one prompt field at a time.

{
  "task": "prompt_revision",
  "previous_issue": "[blurry subject / wrong crop / extra objects / unsafe brand cue / text zone missing]",
  "keep_from_previous": ["[successful element 1]", "[successful element 2]"],
  "revise_only": "[specific field to change]",
  "new_instruction": "[clear replacement instruction]",
  "constraints": ["do not add unsafe workaround", "do not imitate protected brands", "do not invent proof or approval"],
  "output_check": ["issue improved", "kept fields remained stable", "no new safety risk introduced"]
}

It avoids rewriting the whole prompt after every failed result and makes iteration more measurable.