--- name: reversa-image-prompt-json license: MIT compatibility: Claude Code, Codex, Cursor, Gemini CLI, and other Agent Skills-compatible agents. metadata: author: pnocera version: "1.0.0" framework: reversa team: shared-skills role: image-prompt-builder description: > Creates structured JSON prompts for high-quality image generation with luxurious and cinematic aesthetics. Use this skill whenever the user wants to generate an image prompt, create a product photo, build a prompt for an AI image model, virtually photograph a product, create an image of food, beverage, cosmetic, jewelry, fashion, or any visual item. Also activate when the user mentions: "image prompt", "generate product image", "AI product photo", "prompt for Midjourney/DALL-E/Flux", "photograph product", or asks to "build a JSON image prompt". --- # Image Prompt Builder Skill for building structured JSON prompts for product image generation with cinematic and luxurious aesthetics — optimized for **Nano Banana 2 (Gemini 3.1 Flash Image)** via **Google Antigravity**, with support for all native model parameters. --- ## Mandatory workflow When activated, this skill must **ALWAYS** follow these steps in order: 1. **Guided interview** — Collect user information in blocks 2. **Confirmation** — Show summary and request approval 3. **JSON generation** — Build the final structured prompt --- ## STEP 1 — Guided interview by blocks Collect information in **3 rounds of questions**, never all at once. --- ### Round 1 — Product and Scene Ask the user: > "Let's build your image prompt! I need to understand the product first. Tell me:" 1. **Product type**: What is the product? (e.g.: chocolate cake, perfume bottle, sneakers, shake, jewelry...) 2. **Brand name**: Is there a visible brand? If so, what is the name? 3. **Product appearance**: Describe the color, texture, finish, shape. The more detail, the better. 4. **Extra elements**: Are there accompaniments? (fruits, ice, flowers, leaves, reflections...) 5. **Scene type**: What is the overall mood of the image? - Suggested options: luxurious and cinematic / clean and minimalist / dramatic and high-contrast / warm and cozy / futuristic and technological --- ### Round 2 — Composition and Action > "Great! Now tell me about the dynamic visuals of the image:" 6. **Main action**: Is the product static or in motion? (e.g.: liquid exploding, suspended particles, smoke, splash, cut revealing interior...) 7. **Elements suspended in the air**: What elements float around the product? (e.g.: droplets, powder, fragments, leaves, crystals, bubbles...) 8. **Support surface**: Where is the product? (e.g.: polished white marble, matte black stone, rustic wood, transparent glass, abstract surface...) 9. **Camera angle**: How does the camera frame the product? - Options: low angle (dominance) / eye level / slightly above / extreme macro / 3/4 angle --- ### Round 3 — Lighting, Colors, and Technical Specifications > "Almost there! Now the visual and technical part:" 10. **Lighting style**: How do you want the light? - Options: clean and bright studio / dramatic with shadows / soft natural light / luxury product light with rim light / colored neon light 11. **Background color palette**: Which color/gradient dominates the background? (e.g.: charcoal black with amber bokeh, pink to champagne gradient, dark blue to white...) 12. **Accent colors**: What colors appear in the surrounding elements? (e.g.: gold, silver, bright red, pastel tones...) 13. **Resolution**: What quality level do you need? - `512px` — quick iteration / tests - `1K` — social media and digital use - `2K` — professional content - `4K` — maximum production / print 14. **Aspect Ratio**: What image proportion? (default: `16:9`) - `16:9` — widescreen (default) - `1:1` — square (Instagram feed) - `9:16` — vertical (Stories, Reels, TikTok) - `4:3` — classic - `3:4` — portrait - `4:1` / `1:4` — horizontal / vertical banner - `8:1` / `1:8` — super banner 15. **Rendering style**: Ultra-detailed photorealistic / illustration / 3D render / analog photo / other? 16. **Anything else?**: Any special detail you want to ensure in the image? --- ## STEP 2 — Confirmation After collecting all answers, show a **bullet-point summary** for the user to confirm: ``` PROMPT SUMMARY: - Product: [type] — [brand] - Appearance: [description] - Scene: [type] - Action: [description] - Suspended elements: [list] - Surface: [description] - Angle: [angle] - Lighting: [style] - Background: [colors] - Accents: [colors] - Resolution: [e.g.: 2K] - Aspect Ratio: [e.g.: 1:1] - Rendering: [e.g.: ultra-photorealistic] Is this correct? Can I build the JSON prompt now? ``` Only proceed to Step 3 after user confirmation. --- ## STEP 3 — JSON Generation With the confirmed answers, build the prompt following **exactly** this schema: ```json { "master_prompt": { "scene_type": "[speed/style adjective] [niche] photography", "product": { "type": "[rich, adjective-heavy product description]", "brand_name": "[brand name or 'no visible branding']", "appearance": "[detailed color, texture, shape, finish]", "accompaniments": [ "[element 1 with sensory description]", "[element 2 with sensory description]" ] }, "composition": { "action": "[central dramatic action captured in motion]", "surrounding_elements": [ "[suspended element 1 with motion detail]", "[suspended element 2 with motion detail]", "[suspended element 3 with motion detail]" ], "placement": "[centered hero placement on specified surface]" }, "lighting": { "style": "[complete lighting style]", "effects": [ "[rim light effect]", "[key light effect]", "[backlight or top light effect]", "[extra effect if needed]" ] }, "color_palette": { "background": "[background gradient/bokeh with transition description]", "accents": "[comma-separated list of accent colors]" }, "technical_specs": { "camera": "[lens type], [chosen angle]", "shutter": "[capture type — freeze-motion, long exposure, etc.]", "depth_of_field": "[main focus], [blur description]", "rendering_style": "[photorealistic / illustration / 3D render / analog photo / etc.]" }, "output_specs": { "resolution": "[512px | 1K | 2K | 4K]", "aspect_ratio": "16:9", "model": "nano-banana-2", "synthid_watermark": true } } } ``` --- ## JSON quality rules - **Luxury and premium adjectives** are mandatory in every descriptive field - **Frozen motion** must always be present in `action` and `surrounding_elements` - **Reflective surfaces** must be mentioned in `placement` - The product is always the **centered hero** of the scene - `surrounding_elements` must have **minimum 3, maximum 6 items** - `lighting.effects` must always have **3 or 4 effects** (rim, key, back/top + optional extra) - `scene_type` must follow the pattern: `"[speed/style adjective] [niche] photography"` - `output_specs.resolution` must use native Nano Banana 2 values: `512px`, `1K`, `2K`, or `4K` - `output_specs.aspect_ratio` must use natively supported model values - `output_specs.model` must always be `"nano-banana-2"` - `output_specs.synthid_watermark` must always be `true` (mandatory Google default) --- ## After generating the JSON Present the formatted JSON in a code block and add: > **Antigravity usage tip:** Paste this JSON directly into the Nano Banana 2 prompt field in Google Antigravity. The `output_specs` fields are interpreted natively by the model — no additional prefix is needed. Ask if the user wants to adjust any field, change the aspect ratio, or generate variations. --- ## Reference examples For language pattern inspiration, consult `/mnt/skills/user/image-prompt-builder/references/examples.md` if available.