Prompt Optimization Guide

This guide is about prompt quality, not backend flow.

A practical prompt structure

Use this order:

1. Subject
2. Material or visual traits
3. Lighting
4. Composition
5. Mood or usage
6. Negative prompt

Example:

Luxury skincare bottle, frosted glass and brushed aluminum cap, soft front key light, centered composition with negative space on the right, clean premium product photography, no people, no extra props, no text

How prompt strategy changes by model

Flux Schnell

Keep prompts short.

• one subject
• one style direction
• one lighting idea

Best for fast iteration, not maximum nuance.

Qwen Image

Write longer and more structured prompts.

Use explicit sections such as:

• subject
• environment
• lighting
• composition
• mood

This model benefits from clearer instruction blocks.

Nano Banana Pro

Lead with materials and surfaces.

Examples:

• clear glass
• brushed steel
• matte stone
• seamless gray backdrop

This usually improves product-photography control.

Seedream 4

State detail intent directly.

Examples:

• crisp texture
• sharp edge definition
• high-detail
• 4K-ready

This helps communicate the quality target.

Flux Dev / Flux 2 Pro / Flux 2 Max

Write prompts that describe:

• subject consistency
• finish level
• material realism
• light ratio

These models respond better when you are specific about final-image quality.

Prompt optimization rules that usually help

1. Put the subject first

Do not start with style words alone.

Bad:

cinematic premium aesthetic, beautiful composition, sharp and detailed

Better:

Mechanical watch on black stone, cinematic side light, premium macro product photography, sharp dial texture

2. Keep negative prompts focused

Good negatives remove obvious failure modes:

• no people
• no watermark
• no extra props
• no text

Do not turn the negative prompt into a second full prompt.

3. Use composition keywords

Useful words:

• close-up
• top-down
• centered composition
• negative space
• shallow depth of field
• left-aligned subject

4. Match prompt length to model intent

• Flux Schnell: short
• Qwen Image: long and structured
• Nano Banana Pro: product-first and material-first
• Seedream 4: detail-first
• Flux 2 Pro / Max: finish-first and quality-first

Search-friendly prompt keywords

If you want the page and the image output to align better, combine goal words with visual control words:

• product photo, hero shot, still life, tabletop, commercial lighting
• high detail, sharp texture, premium finish, clean background
• negative space, centered composition, left-aligned subject, close-up
• no clutter, no people, no watermark, no extra props

Prompt goal examples

• product hero image
• high-detail watch render
• still life tabletop scene
• clean ecommerce banner
• bilingual composition-controlled prompt

Next steps