By name: ai-artist description: Generate images via Nano Banana with curated prompts category: AI Generation & Multimodal status: active
AI Artist
Purpose
Generate high-quality images using Google's Nano Banana image models (Gemini API) with prompt engineering best practices. Leverage reference images, specific photographic language, and composition cues to produce consistent, professional visuals.
When to Use
- Marketing and product photography (mockups, concept art)
- Social media content and blog headers
- UI/UX design mockups and placeholder imagery
- Icon and illustration generation
- Style exploration and rapid visual iteration
- Batch image generation with reference consistency (up to 14 reference images per request)
Do NOT use when: Legal/copyright-sensitive content, deepfakes, or images requiring pixel-perfect technical precision.
Workflow
Define Visual Scope — Establish subject, setting, lighting, composition style, and photographic language.
Craft Detailed Prompt — Include subject (what), environment (where), lighting (how), composition (perspective), and style (aesthetic). Use 50–100+ words.
Add Reference Images — (Optional) Provide up to 14 reference images for style transfer, character consistency, or compositional guidance. Nano Banana uses these to inform output.
Generate & Iterate — Submit prompt. Review output. Refine prompt based on results (adjust lighting, composition, style) and regenerate.
Post-Process — Export images for design use. Integrate into mockups, marketing collateral, or UI designs.
Key Concepts
Photographic & Cinematic Language
Instead of abstract descriptions, use professional photography terminology. E.g., instead of "a mug," write "A photorealistic product photo of a ceramic coffee mug, soft natural lighting from the left, white background, sharp focus, commercial photography style, 4K detail." Terms like "wide-angle shot," "macro shot," "low-angle perspective," "85mm portrait lens," "Dutch angle," "shallow depth of field" translate into visual composition cues.
Prompt Structure
Best results combine: Subject (what/who) + Environment (where/context) + Lighting (how it's lit, time of day, mood) + Composition (framing, perspective, focal point) + Style (photographic, cinematic, painting, illustration, art style).
Reference Images for Consistency
Nano Banana accepts up to 14 reference images per request. Use them for: style transfer (apply a reference painting's aesthetic to new subjects), character consistency (same person across multiple images), or compositional guidance (match framing and perspective).
Model Selection
- Nano Banana 2 (Feb 2026): High efficiency, lower price ($0.02/image), ideal for high-volume batch generation.
- Nano Banana Pro: Highest quality ($0.06/image), best for hero images and critical visuals.
- Imagen 4 tiers: Premium options for ultra-high fidelity.
Thinking Level & Advanced Features
Gemini 3 models use dynamic thinking by default, allowing the model to reason through complex prompts before generating images. Control reasoning depth with the thinking_level parameter for balance between speed and accuracy.
Example
Prompt (Weak): "A futuristic city at night"
Prompt (Strong): "A futuristic cyberpunk megacity at night, neon-lit skyscrapers piercing through smog, flying vehicles with glowing trails, seen from a low-angle wide-angle perspective (24mm lens), neon blues and purples dominant, rain-slicked streets reflecting neon lights, cinematic lighting with volumetric fog, shot in the style of Blade Runner 2049, ultra-detailed, 8K, photorealistic."
With Reference: Add 3 reference images: one cyberpunk aesthetic, one night city lighting, one neon color palette. Nano Banana uses these to guide generation.
Output: Consistent, detailed, cohesive cyberpunk scene suitable for concept art, marketing, or game design.
Common Pitfalls
- Vague prompts: "A beautiful landscape" produces mediocre results. Be specific: terrain, season, time of day, weather, perspective.
- Ignoring photographic language: Abstract descriptions underutilize the model. Use lens terms, lighting positions, focal lengths.
- No reference images: Missing style consistency across batches. Provide 1–3 key references.
- Overloaded prompts: 200+ words of contradictions confuse the model. Prioritize 5–10 key attributes.
- Ignoring thinking_level: Fast generation (low thinking) sacrifices quality. Balance with project needs.
- Not iterating: First output is rarely perfect. Use results to refine and regenerate.
References
- Google Gemini Image Generation API — Official API docs for Nano Banana and Imagen models
- Nano Banana Prompting Guide — Google Cloud's comprehensive prompt engineering strategies
- Gemini Prompt Design Strategies — Official best practices for effective prompts
- Gemini Image API 2026 Guide — Updated pricing, model selection, and advanced features