Ecom Shot 📸

Step 2: Confirm Subject & Gather Preferences

Use ask_user_input_v0 to collect the user's preferences in one go:

Question 1 — Subject Confirmation (single_select): Present what you identified as the main subject and ask the user to confirm or correct.

• Options: your identified subject description, or "其他（我来描述）"

Question 2 — Scene Type (single_select):

• 纯白底主图（White Background Hero Shot）
• 生活场景图（Lifestyle Shot）
• 创意氛围图（Creative Atmosphere）
• 自定义场景（Let me describe）

Question 3 — Aspect Ratio (single_select):

• 1:1（默认，适合淘宝/Amazon）
• 4:3（横版展示）
• 3:4（竖版/小红书）
• 16:9（Banner/横幅）

Step 3: Generate the Prompt

Read the reference file at references/prompt-templates.md for scene-specific templates and photography parameters, then compose the final prompt.

Prompt Composition Principles

Nano Banana Pro is a "Thinking" model built on Gemini 3 Pro. It understands intent, physics, and composition. The key principles:

1. Think like a Creative Director, not tag soup. Write in natural, descriptive English sentences — not comma-separated keyword lists. Nano Banana Pro's text encoder is an LLM that reasons about the prompt before generating.

2. 6-element structure — Every prompt should include:

   • Subject: Precise description of the product with material, color, and distinguishing features
   • Composition: Camera angle, framing, how the product fills the frame (~85-90%)
   • Camera & Lens: Focal length (e.g., 85mm), aperture (e.g., f/8), ISO
   • Lighting: Specific setup (softbox, key light angle, fill ratio, rim light)
   • Style: The overall aesthetic (commercial product photography, editorial, etc.)
   • Technical specs: Resolution and aspect ratio

3. Reference image directive — Always start with a clear instruction to use the uploaded image as reference: "Using the provided reference image as the product source, ..."

4. Negative instructions — Include what to exclude: no text, no watermarks, no hands, no distracting elements.

5. Resolution & Aspect Ratio — Specify the exact output: "Generate at 2K resolution (2048×2048 for 1:1)" or the appropriate dimensions for the chosen ratio.

Output Format

Present the prompt in a clean code block that the user can copy directly:

📸 Ecom Shot — {Product Name}
━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Scene: {scene type}
Aspect Ratio: {ratio}
Resolution: 2K

--- Prompt (copy below) ---

{the generated prompt text}

--- End Prompt ---

After the prompt, add a brief "💡 Tips" section with 1-2 actionable suggestions:

• How to iterate if the result isn't perfect (e.g., "If the lighting feels flat, try adding: 'with dramatic side lighting creating deep shadows'")
• Whether multi-turn editing could help (e.g., "After generating, you can ask Gemini to 'make the background slightly warmer'")

Step 4: Iterate if Needed

If the user wants to adjust the prompt:

• For small tweaks: modify inline and re-output
• For a different scene type: go back to Step 2
• For additional variants: generate multiple prompts (e.g., white background + lifestyle pair)

Key Rules

• Prompts are ALWAYS in English — Nano Banana Pro performs best with English prompts, even for Chinese-market e-commerce images
• Analyze the actual image carefully — Don't guess the product. Look at material, color, shape, brand details
• Be specific about camera parameters — Use real photography terminology (focal length, f-stop, lighting rigs)
• Default to 1:1 at 2K unless the user specifies otherwise
• No text in the generated image — E-commerce hero shots should be text-free; add text overlays separately
• Preserve product identity — The prompt must instruct Nano Banana to maintain the exact appearance of the product from the reference image
• Conversation and explanations should use the user's preferred language (check memory), but the prompt itself is always English
• Use Chinese full-width punctuation for Chinese conversation content