Prompt Writing Basics

What a good prompt needs

# A good AI generation prompt does not need to be long. It needs to be clear about what should appear, how it should look, and what constraints must stay fixed.

For KrafLayer, a strong prompt usually includes five parts: subject, visual style, composition, lighting, and output intent. If you are generating a product image, add product clarity and background direction. If you are generating video, add motion and camera behavior.

A simple structure is:

• Subject: what the image or video is about.
• Style: photoreal, anime, cinematic, product photography, editorial, illustration, or another visual direction.
• Composition: close-up, full body, flat lay, hero shot, centered product, wide shot, split layout, or grid.
• Lighting and mood: soft daylight, studio lighting, neon, warm interior, dramatic backlight, clean commercial light.
• Constraints: aspect ratio, number of subjects, reference preservation, brand colors, readable text, or background requirements.

Weak prompt:

A nice product photo of a bottle.

Stronger prompt:

A clean commercial product photo of a matte white skincare bottle on a pale green acrylic surface, centered hero composition, soft studio lighting, subtle water reflections, minimal premium beauty brand style, sharp product label, 1:1 aspect ratio.

How KrafLayer prompt enhancement works

KrafLayer's prompt enhancer is designed to turn a rough idea into a model-ready prompt without changing the user's core intent.

It keeps important constraints such as numbers, layout words, aspect ratio, reference-image instructions, product identity, and requested style. If the user writes in Chinese, the enhanced prompt stays in Chinese. If the user writes in English, it stays in English.

The enhancer can also adapt the prompt based on context:

• Image generation prompts focus on subject, composition, visual style, lighting, detail, and output quality.
• Product prompts prioritize clarity, material realism, commercial composition, and clean background control.
• Reference-image edits preserve the identity of the input image and describe what should change.
• Video prompts add motion, camera movement, pacing, scene continuity, and first-frame or end-frame guidance.
• Style presets can bias the result toward cinematic, commercial, illustration, social, anime, fashion, or product directions.

Start with the subject

The subject is the anchor of the prompt. It tells the model what must appear.

For characters, describe identity, outfit, pose, expression, and role.

Example:

A game-style anime female sword fighter, silver armor, short black hair, confident expression, holding a glowing blue blade, full-body character design.

For products, describe the product category, material, color, shape, and brand mood.

Example:

A premium wireless speaker with a soft fabric grille, rounded aluminum body, matte charcoal finish, placed on a clean studio surface.

For environments, describe the location, time, atmosphere, and important objects.

Example:

A compact creative studio at night, desk with drawing tablet, warm lamp light, wall of reference images, quiet cinematic mood.

Add style without overloading the prompt

Style helps the model choose the visual language. Use a few precise style terms instead of stacking too many unrelated tags.

Good style signals include:

• photoreal commercial product photography
• cinematic science fiction concept art
• Japanese RPG character design
• clean editorial fashion photography
• cozy lifestyle interior photography
• polished 3D icon style
• watercolor storybook illustration

Avoid mixing style directions that fight each other, such as photoreal product photography, watercolor, clay render, cyberpunk, vintage film, and flat vector in the same prompt unless you intentionally want a hybrid look.

Describe composition and camera

Composition controls where things appear in the frame. This matters especially for product images, thumbnails, posters, and ecommerce visuals.

Useful composition phrases:

• centered hero shot
• full-body character pose
• close-up portrait
• wide establishing shot
• top-down flat lay
• three-quarter product view
• symmetrical composition
• negative space for headline text
• clean background with product centered
• grid of three variations

Camera phrases help when you want a specific visual feel:

• macro lens
• 35mm editorial lens
• shallow depth of field
• low-angle heroic shot
• eye-level product view
• overhead camera
• slow push-in camera movement for video

Use lighting as a quality control

Lighting often changes the perceived quality of the result more than extra detail words.

For clean commercial images, use:

• soft studio lighting
• diffused daylight
• controlled highlights
• gentle shadow under the product
• bright ecommerce background

For cinematic images, use:

• dramatic rim light
• volumetric light
• warm practical lights
• neon reflections
• moonlit backlight

For anime or game character art, use:

• clean key light
• glowing weapon light
• atmospheric background light
• high-contrast character silhouette

Preserve hard constraints

Hard constraints are details that should not change. Put them clearly in the prompt.

Examples:

• exactly three characters
• 1:1 aspect ratio
• no text on the image
• keep the same product shape
• preserve the face identity from the reference image
• white background
• front view only
• include one hero image and four detail images

KrafLayer's prompt enhancer tries to preserve hard constraints, but you should still keep the most important ones explicit.

Prompting with reference images

When using a reference image, describe what should stay the same and what should change.

Good reference-image edit prompt:

Use the uploaded product as the exact reference. Preserve the product shape, color, label placement, and material. Replace the background with a clean beige studio setting, add soft commercial lighting, and create a premium ecommerce hero image.

For character or portrait references:

Use the reference image to preserve the subject's identity, facial structure, hairstyle, and outfit silhouette. Transform the scene into a cinematic fantasy character portrait with dramatic rim lighting and a dark forest background.

Avoid vague edit prompts like:

Make it better.

Instead, say what better means: cleaner background, sharper product edges, more realistic lighting, stronger composition, different style, or higher commercial polish.

Prompting for product image sets

A product image set should not be one prompt repeated several times. Each image should explain a different selling angle.

A strong five-image product set might include:

• Hero image: clean product-first composition.
• Lifestyle scene: product used in a realistic environment.
• Detail shot: material, texture, buttons, label, or craftsmanship.
• Feature scene: visual explanation of the key benefit.
• Social ad image: more expressive composition with space for copy.

Example:

Generate a five-image product set for a minimalist smart water bottle. Include a clean ecommerce hero shot, a gym lifestyle scene, a close-up of the lid sensor, a hydration reminder feature scene, and a social ad image with fresh blue lighting.

This is the reason KrafLayer's agent workflow may create several different prompts for one product request instead of running the same prompt multiple times.

Prompting for AI video

Video prompts need motion. A still-image prompt describes a frame; a video prompt describes change over time.

Add:

• subject action
• camera movement
• environment motion
• pacing
• start and end state
• mood continuity

Example image prompt:

A cinematic product shot of a luxury perfume bottle on black glass, dramatic reflections, soft gold backlight.

Example video prompt:

A cinematic 6-second product video of a luxury perfume bottle on black glass. The camera slowly pushes in from a three-quarter angle while gold light moves across the bottle, subtle mist drifts behind it, reflections shimmer on the surface, premium fragrance ad mood, smooth motion, no text.

If you use image-to-video, the prompt should respect the starting image:

Animate the uploaded product image. Keep the bottle shape, label, color, and composition consistent. Add a slow camera push-in, soft moving reflections, and subtle background mist. Do not change the product identity.

When to use negative prompts

Negative prompts are useful when a model supports them, but they should be short.

Common negative prompt examples:

• blurry
• low quality
• distorted hands
• extra fingers
• unreadable text
• watermark
• messy background
• duplicated product
• deformed face

Do not put your main creative direction in the negative prompt. Use it only to reduce common failure modes.

Before generating, check the prompt

Before you spend credits, check four things:

• Is the subject clear?
• Is the style specific?
• Are the must-keep constraints explicit?
• Does the prompt match the selected model and format?

For product images, also check whether the prompt protects the product identity.

For video, check whether the prompt includes motion, camera behavior, and duration-friendly action.

Example prompt templates

Product hero image

A clean commercial product photo of [product], [material/color details], centered hero composition, [background], soft studio lighting, realistic shadows, premium ecommerce style, sharp product edges, [aspect ratio].

Anime game character

A game-style anime character design of [character role], [outfit and weapon], [pose], [expression], [background setting], clean character silhouette, detailed costume design, polished RPG key art style, [aspect ratio].

Cinematic image

A cinematic image of [subject] in [environment], [camera angle], [lighting], [mood], detailed atmosphere, realistic depth, high-quality visual composition, [aspect ratio].

Image-to-video

Animate the uploaded image while preserving [subject/product identity]. Add [subject motion], [camera movement], [environment motion], [lighting change], smooth pacing, consistent details, no unwanted transformation.

Related KrafLayer docs

• [Image generation models](/docs/image-generation)
• [Video generation models](/docs/video-generation)
• [AI image-to-video guide](/docs/ai-image-to-video-guide)
• [Generation cost](/docs/generation-cost)