Prompting tips for AI Image Models

When choosing an AI image model, the right option depends on your creative goals, the level of detail required, and the desired speed of results. Some models excel at clean, consistent design for branding and graphics, while others specialize in photorealism, high-speed drafts, or playful transformations. Below is a guide on when to reach for different models—along with prompting tips to get the most out of each.

Check it out

Seedream 4.0

Best for: High-res deliverables, multi-image consistency, on-image text. Examples include branding, campaign storytelling, and artistic campaigns

Use it when you need:

• Consistent product/character across multiple shots
  • It can produce batches or a series of images that maintain consistency (style, character features etc
• Style transfer or fusion from multiple reference images
  • Able to merge or fuse aspects from multiple images (e.g. combining elements, blending styles, placing a character from one into a scene of another) coherently
• To add/remove objects while preserving visual accuracy
• Precise on-image text

Seedream 4.5

Best for: Seedream 4.5 is best for producing cinematic, high-quality visuals with strong consistency, precise instruction following, realistic spatial layouts, and knowledgeable, technically accurate imagery—ideal for professional work in e-commerce, film, advertising, gaming, education, and design.

Use it when you need: 

• Superior Aesthetics: Produce cinematic visuals with refined lighting and rendering.
• Higher Consistency: Maintain stable subjects, clear details, and coherent scenes across multiple images.
• Smarter Instruction Following: Accurately responds to complex prompts with precise visual control and interactive editing.
• Stronger Spatial Understanding: Generate realistic proportions, object placement, and scene layout.
• Richer World Knowledge: Create knowledge-based visuals with accurate scientific and technical reasoning.

Prompting tips:

1. Be explicit about what to preserve and clearly define image editing goals

   Recommended: “Keep the same character face and outfit, but change the background to a forest at sunset.
   Avoid: "Make the background a forest"

2. Use multiple reference images to lock in style + identity

3. For text: put desired wording clearly in double quotation marks

   Recommended: “Write “WELCOME HOME” on the billboard in bold white letters.”
   Avoid: "Generate a Billboard with the words welcome home."

4. Describe edits via plain language

   Recommended: “Add or remove objects”, “change style”, “adjust backgrounds” “change lighting”
   Avoid: "Make it better", "fix this"

5. Specify the image purpose and type in your prompt

   Recommended: "Design a logo for a music streaming company. The logo features a cat wearing headphones and sitting on a vinyl record. The company name "MEOWTUNES" is written on it."
   Avoid: An abstract image with a cat sitting on a vinyl record, and the word MEOWTUNES on it."

Nano Banana 

Best for: Speed, casual/social use, fun edits. Examples include quick marketing concepts, social media visuals, and small assets.

Use Nano Banana when you need:

• Quick, polished edits for social content, memes, playful transforms
  • Take a single photo and make many faithful variants — change style, background, or theme while keeping the person/object recognizable.
• Style transfer & playful transformations
  • It shines at turning photos into stylized outputs (miniature figurines, watercolor, video-game skins, fashion remixes) that are shareable and visually distinct
• Conversational, iterative editing
  • You can ask multiple sequential edits in plain language (e.g., “make it sunset,” then “change dress color to emerald”) and the model preserves context across turns. 

Use Nano Banana Pro when you need: 

• Improved consistency:  It maintains the highest subject consistency so you can generate the same character’s face, hairstyle or expression across multiple images or edits.
• Text rendering: It generates crystal-clear text in different languages when clear about the text, the font style (descriptively), and the overall design.
• High-quality informational images: It can easily generate informational images like diagrams or infographics.
• Real-world knowledge: It has a wider range of knowledge, generating images that are true to life. 

Prompting tips:

1. Keep requests short and descriptive (Nano Banana is tuned for casual use)

2. Works best with single reference image edits — upload photo, then describe edit

3. Expect speed and fun, not extreme photorealism in tricky lighting/materials

Prompt example:Use the woman and dog sitting on the snow from the snowy field image, make the horizon transform from snow into the tropical sea from the beach image, add the bare tree from the starry night tree image beside them.

Original images

Generated image

Prompt example: create a photo with this couple standing in this field in the middle left of the field and an airplane in the sky right above the couple

Original Images

Generated Image

Prompt example: Remove the blurriness from the image, making the image very clear.

Original Image

Generated Image

Nano Banana Pro

Best for: Speed and higher-quality stylized edits, making faithful variants of the same image, style transfers (e.g., clay, watercolor, diorama, toy aesthetics), and conversational editing across multiple turns

Use it when you need:

• High-Fidelity text rendering: Accurately generate images that contain legible and well-placed text, ideal for logos, diagrams, and posters.
• Standout Stylizations: Stylize your images to perfection, with styles such as clay figurines, watercolor paintings, video-game or animated skins, oy, diorama, and miniature looks, and fun fashion reworks.
• Conversational Editing: You can stack edits naturally. Edits maintain context, allowing smooth, iterative refinement.
• Character consistency: Build on character consistency and scene description to create panels for visual storytelling.

Prompting tips:

1. Use photography terms, for realistic images. Mention camera angles, lens types, lighting, and fine details to guide the model toward a photorealistic result.

   Recommended: “Create a realistic portrait: soft natural window light, 50mm lens, shallow depth of field.”
   Avoid: “Fix all lighting to look real.”

2. To create stickers, icons, or assets, be explicit about the style and request a transparent background.

   Recommended: “Create a cute sticker-style illustration of this dog: thick outlines, soft pastel colors, simplified shapes, and a transparent background.”
   Avoid: “Turn this into an illustration with no background.”

3. Be very clear about the wording, the font style (descriptive terms only), and the design mood.

   Recommended: “Add the text ‘Fresh Start’ in large, rounded sans-serif lettering at the top. Use a clean, modern layout with soft shadows and plenty of spacing.”
   Avoid: “Give it a cool font.”

4. Be Hyper-Specific. The more detail you provide, the more control you have. 

   Recommended: “Ornate elven plate armor, etched with silver leaf patterns, with a high collar and pauldrons shaped like falcon wings."
   Avoid: “Fantasy armor" 

Flux Kontext

Best for: Iterative editing, reproducible pipelines, design/development workflows. Examples include product updates, iterative design work, or controlled revisions.

Use it when you need:

• Incremental edits where untouched areas must remain stable (design versions, character continuity).
• Complex multi-round workflows
• Complex edits that need to hold context across multiple rounds (e.g. same room, same character, different lighting or outfits)

Prompting tips:

1. Give clear local edit instructions

   Recommended: “Only change the color of the chair to red, leave the rest of the room identical.”
   Avoid: "Change the chair"

2. Use for progressive refinements rather than one-shot perfection

   Recommended: Broad scene → Details → Polish

   Broad scene (first pass):
   “Add a mountain range in the background of the landscape.”

   Details (second pass):
   “Add a small cabin near the trees on the right side.”

   Polish (final pass):
   “Make the sky slightly pink at sunset, leave everything else the same.”

   Avoid: “Create a perfect, photorealistic mountain landscape with detailed mountains, a cabin by the trees, smoke from the chimney, and a glowing pink sunset sky — everything should be flawless.”

3. Stress consistency across edits

   Recommended: “Keep character face identical in all frames.”
   Avoid: "Make sure everything stays the same."

Flux 2.0 Pro

Best for: Users who need top-tier, production-grade image quality with exceptional detail, realism, and prompt accuracy. It’s best suited for commercial work, client-ready visuals, product imagery, marketing assets, photorealistic renders, and any scenario where output polish and reliability are critical. Examples include commercial work, client-facing deliverables, product imagery, marketing assets, photorealistic renders, and any project where the final output must meet high professional standards.

Use it when you need:

• Highest fidelity results with rich texture, accurate lighting, and clean details
• Precise prompt following for complex scenes, compositions, or technical requirements
• Consistent outputs across multiple generations or variations
• Professional-quality visuals for print, advertising, video storyboards, or concept development
• Text rendering that is clean and legible, suitable for packaging, posters, and UI mockups

Flux 2.0 Dev

Best for: Experimentation, prototyping, fine-tuning, custom pipelines, and multi-image editing where flexibility and control matter more than maximum polish.

Use it when you need:

• Multi-reference editing to maintain consistent characters, styles, or objects across multiple images
• Rapid iteration or prototyping, especially when exploring variations or creative concepts.

Prompting tips:

1. Start with clear subject + action + setting to help Flux 2.0 lock onto your intent.

   Recommended: “Golden retriever puppy splashing in a forest creek at sunrise.” 
   Avoid: “A cute scene with a dog doing something in nature.”

2. Add style keywords (e.g., “hyper-realistic,” “studio lighting,” “flat illustration,” “cinematic”) to shape aesthetics.

   Recommended: “Portrait of an astronaut, cinematic lighting, hyper-realistic texture.” 
   Avoid:“A picture of an astronaut, make it look cool.”

3. If you want a specific composition, mention camera angle, framing, and mood.

   Recommended:  “Low-angle shot of a skyscraper rooftop garden, wide frame, calm and atmospheric.”
   Avoid: “A rooftop garden but make it interesting.”

4. Keep it concise — Flux 2.0 responds well to compact, structured prompts rather than long paragraphs.

   Recommended: “Mountain cabin in winter, smoke from chimney, soft overcast light.” 
   Avoid: “A detailed image showing a cabin somewhere in the mountains during a cold season, with lots of snow everywhere and maybe smoke

Kling O1 Image

Best for: Consistent long-term character or product shots

Use it when you need:

• Enhanced continuity: The model understands inputs better (especially with the "@" syntax), it maintains character identity and object permanence far more effectively.
• Sharper texture retention: Expect less "AI shimmer" or boiling artifacts in background details.
• Consistency: Character consistency that rivals the stability we’ve seen in static image models.

Prompting tips:

1. Upload reference images and simultaneously specify changes like scenery, color, material, and style.

2. Use native reference tagging to organize your reference images.

   Recommended:  “Put the headphones from @image1 on the character from @image2.” 
   Avoid: Put the headphones on the character from the image.

3. Skill combos (Stacking Tasks): Because O1 is a unified model, you can “stack” operations in a single pass to save time and credits. 

   Recommended: "Add a subject walking into the frame AND change the location to a snowy street" 
   Avoid: First iteration -  “Add a subject walking into the frame”. Second iteration - “Change the location to a snowy street.”

4. Refine with natural language.

   If you are editing, simply type your change: "Change the background to a cyberpunk city."

Try it now

Need some more help? Head back to our Help Center.