GPT-4o Image: Native Multimodal Creation & Editing
Experience the power of GPT-4o for image generation and editing! Create stunning visuals, refine images with maskless edits, and leverage reference images, all within a natively multimodal system.
Overview
GPT-4o Image Generation, powered by OpenAI’s GPT-4o model, is a groundbreaking tool that integrates text and image processing into a single, natively multimodal system. This allows for unprecedented flexibility in creating and editing visuals. ImagenCraft provides a platform to utilize GPT-4o’s capabilities, excelling at accurate text rendering, precise prompt following, and leveraging in-context learning from uploaded images. It offers distinct Generate and Edit modes to cover a wide range of creative and practical applications.Natively Multimodal
Integrates text and image processing in one system for unprecedented flexibility.
Accurate Text Rendering
Seamlessly integrates clear and precise text into generated images.
Precise Prompt Following
Excels at adhering to detailed instructions and handling multiple objects.
In-Context Learning
Leverages uploaded images as visual context and inspiration.
Modes of Operation
GPT-4o Image offers two primary modes to suit your creative workflow:Generate Mode: Create Images from Text
In Generate mode, you can create entirely new images from scratch by providing a text prompt.
- Prompt: Your text description of the desired image. (Max 1000 characters)
- Quality: Control the level of detail and generation time.
- Possible values:
low(Faster),medium,high(Slower),auto.
- Possible values:
- Size: Select the output resolution and aspect ratio.
- Possible values:
1024x1024(Square),1024x1536(Portrait),1536x1024(Landscape),auto.
- Possible values:
- Background: Choose the background type.
- Possible values:
opaque,transparent(Note: Availability depends on Output Format).
- Possible values:
- Moderation: Adjust content filtering sensitivity.
- Possible values:
auto(Standard),low(Less Restrictive).
- Possible values:
- Output Format: File format (PNG, JPEG, WebP).
- Compression Quality: For JPEG/WebP output (0-100).
Generate Mode Examples
Explore the diverse range of images you can generate from text prompts:Sustainable Campaign
Skincare Thumbnail
Infographic
Single Page Comic
Edit Mode: Maskless Image Modification
Edit mode allows you to modify an existing image by providing written instructions. It offers different edit types to suit your needs. Requires an Image Upload (except for Reference Images).
Draw Mask
Draw Mask
Modify specific areas by drawing a mask on the uploaded image.
- Inputs: Prompt (describing changes in masked area), Uploaded Image, Mask Image (drawn), Quality, Size, Background, Moderation.
Maskless Editing
Maskless Editing
Modify the image based on a prompt without drawing a mask. The AI determines where to apply changes.
- Inputs: Prompt (describing changes), Uploaded Image, Quality, Size, Background, Moderation.
Reference Images
Reference Images
Generate a new image based on a prompt and up to 4 uploaded reference images. The AI blends concepts from the references.
- Inputs: Prompt (describing how to use references), Reference Images (up to 4), Quality, Size, Background, Moderation.
- Prompt: Describes the desired changes (Maskless, Draw Mask) or how to use references (Reference Images).
- Quality: Control output quality (Same options as Generate).
- Size: Select output size (Same options as Generate).
- Background: Choose background type (Same options as Generate).
- Moderation: Adjust content filtering (Same options as Generate).
- Uploaded Image: The base image to edit (for Draw Mask, Maskless).
- Mask Image: The mask drawn on the uploaded image (for Draw Mask).
- Reference Images: Images to use as visual inspiration (for Reference Images).
Edit Mode Examples
See how GPT-4o Image can modify existing images using its editing capabilities:Maskless Editing Example
Modify an image based on a text prompt without drawing a mask:Input Image
Result
Reference Images Example
Generate a new image by blending concepts from multiple reference images:Reference Image 1
Reference Image 2
.png)
Reference Image 3
.png)
Result
Mastering Prompts for GPT-4o Image
Prompting is crucial in all modes, but the focus shifts depending on whether you are creating, editing, or customizing. GPT-4o models are known for strong prompt interpretation and respond well to descriptive, clear language.Prompt Writing Basics: Subject, Context, and Style
A good starting point for any prompt is to define the core elements:Subject
The main object, person, animal, or scenery.
Context/Background
The environment or setting for the subject.
Style
The artistic style (e.g., painting, photograph, sketch, or more specific styles).
General Prompting Principles:
Be Specific and Detailed
Be Specific and Detailed
Employ descriptive language, detailed adjectives, and adverbs to paint a clear picture for Imagen.
Example: Instead of “a park,” try “A park in the spring next to a lake, the sun sets across the lake, golden hour, red wildflowers.”
Use Natural Language
Use Natural Language
Formulate prompts using descriptive sentences, as you would describe an image to another person.
Consider Negative Prompts
Consider Negative Prompts
Use negative prompts to steer away from unwanted elements (Note: support varies by model).
Prompt Enhancement
Prompt Enhancement
For models supporting prompt enhancement (like
imagen-3.0-generate-002), a shorter prompt can be automatically expanded for potentially better results. This is enabled by default.Prompting for Specific Modes/Tools:
Create Mode
Create Mode
Focus on fully describing the desired image from scratch, combining subject, context, and style with rich details.
Example: “A futuristic cityscape at sunset, high angle view, digital painting, vibrant colors.”
Edit Mode (Draw Mask)
Edit Mode (Draw Mask)
Describe what you want to appear within the masked area. Focus on the object or scene you want to generate and how it should blend with the existing image. For best results, use a description of the masked area. Avoid single-word prompts.
Example: (Mask over a blank wall) Prompt: “A vibrant graffiti mural covering the wall.”
Edit Mode (Maskless Editing)
Edit Mode (Maskless Editing)
Describe the desired changes to the image in natural language. The AI will interpret your instructions and apply the edits without a mask.
Example: (Input Image: Photo of a car) Prompt: “Change the car to red.”
Edit Mode (Reference Images)
Edit Mode (Reference Images)
Describe the desired image, indicating how the AI should draw inspiration from the reference image(s). You can reference specific images if using multiple.
Example: (Reference Image 1: A character, Reference Image 2: A scene) Prompt: “Generate a portrait of the person from image1 standing in the setting from image2.”
Edit Mode (Outpainting)
Edit Mode (Outpainting)
Describe the content you want to appear in the expanded areas around the original image. You can provide an empty string to create the edited images, but a description of the masked area is recommended for best results.
Example: (Outpainting around a portrait) Prompt: “A lush forest extending around the person.”
Edit Mode (Product Image)
Edit Mode (Product Image)
Describe the desired background or environment for the product.
Example: (Input: Product image) Prompt: “Place the product on a wooden table in a sunny cafe.”
Customize Mode (Subject)
Customize Mode (Subject)
Describe the desired image, referencing the subject from your input image. Use the format
[referenceId] to refer to the subject image(s).
Example: (Input: Subject Image with referenceId 1) Prompt: “Generate an image of the person [1] as a knight in shining armor.” (Referencing the subject with [1]).Customize Mode (Style)
Customize Mode (Style)
Describe the desired image content, indicating it should be in the style of your input image. Use the format
[referenceId] to refer to the style image.
Example: (Input: Style Image with referenceId 1) Prompt: “Generate an image of a cat sitting on a chair in the style of image [1].”Customize Mode (ControlNet)
Customize Mode (ControlNet)
Your prompt describes the content and style of the image, while the ControlNet image provides the structural guide. Ensure your prompt aligns with the structural guidance (e.g., edges, pose) provided by the ControlNet image.
Example: (Input: Control Image - Canny edges of a building) Prompt: “A beautiful watercolor painting of an ancient castle at sunset.”
Advanced Prompting Techniques:
Generating Text in Images
Generating Text in Images
GPT-4o can add text to images.
- Iterate: You may need multiple attempts to get the desired text appearance and placement.
- Keep it Short: Limit text to 25 characters or less.
- Multiple Phrases: Experiment with 2-3 distinct phrases.
- Guide Placement: While placement can vary, you can attempt to guide it in the prompt.
- Inspire Font Style/Size: Specify general font styles or size indications (small, medium, large). Example: “A poster with the text ‘Summerland’ in bold font as a title, underneath this text is the slogan ‘Summer never felt so good’.”
Prompt Parameterization
Prompt Parameterization
For API/SDK use, you can parameterize prompts with placeholders like
{logo_style} to be filled by user inputs in an interface.
Example Template: “A logo for a company on a solid color background. Include the text .”Using Specific Styles
Using Specific Styles
Specify artistic styles (e.g., “photography,” “illustration,” “digital art”) or reference historical art movements (“impressionism,” “renaissance,” “pop art”) or specific artists.
Example: “An [art style or creation technique] of an angular sporty electric sedan with skyscrapers in the background.”
Photography Modifiers
Photography Modifiers
Use keywords to influence camera settings and style:
- Camera Proximity: “Close up,” “taken from far away,” “zoomed out.”
- Camera Position: “aerial,” “from below.”
- Lighting: “natural lighting,” “dramatic lighting,” “warm lighting,” “cold lighting,” “studio photo.”
- Camera Settings: “motion blur,” “soft focus,” “bokeh,” “portrait.”
- Lens Types: “35mm,” “50mm,” “fisheye,” “wide angle,” “macro,” “telephoto zoom.”
- Film Types: “black and white,” “polaroid.”
Shapes and Materials
Shapes and Materials
Describe objects made of unusual materials or shapes.
Example: “a duffle bag made of cheese,” “neon tubes in the shape of a bird,” “an armchair made of paper, studio photo, origami style.”
Image Quality Modifiers
Image Quality Modifiers
Use keywords to indicate desired quality level:
- General: “high-quality,” “beautiful,” “stylized.”
- Photos: “4K,” “HDR,” “Studio Photo.”
- Art/Illustration: “by a professional,” “detailed.” Example: “4k HDR beautiful photo of a corn stalk taken by a professional photographer.”
Aspect Ratios and Use Cases
Aspect Ratios and Use Cases
Choose the aspect ratio that best suits your content:
- Square (1:1): General use, social media posts.
- Fullscreen (4:3): TV, media, film, captures more horizontally.
- Portrait full screen (3:4): Fullscreen rotated, captures more vertically.
- Widescreen (16:9): TVs, monitors, mobile (landscape), good for landscapes.
- Portrait (9:16): Widescreen rotated, for tall objects, mobile (portrait), short-form video.
Photorealistic Images
Photorealistic Images
Combine keywords for specific photorealistic subjects:
- Portraits: Prime/zoom lens (24-35mm), “black and white film,” “Film noir,” “Depth of field,” “duotone.”
- Objects (Still Life): Macro lens (60-105mm), “High detail,” “precise focusing,” “controlled lighting.”
- Motion: Telephoto zoom (100-400mm), “Fast shutter speed,” “Action or movement tracking.”
- Wide-angle (Landscape/Astronomical): Wide-angle lens (10-24mm), “Long exposure times,” “sharp focus,” “smooth water or clouds.”
How to Use GPT-4o Image
Navigate through the different modes and tools with this general workflow:Select Your Mode
Choose Generate to create a new image or Edit to modify an existing one.
Select Your Edit Type (Edit Mode)
If in Edit mode, select the type of editing you want to perform: Draw Mask, Maskless Editing, or Reference Images.
Upload Image(s) (If Applicable)
If in Edit mode, upload the necessary base image (Draw Mask, Maskless Editing) or reference images (Reference Images). If using Draw Mask, you will also need to draw a mask on the uploaded image.
Provide Your Prompt
Enter your text description. The prompt’s focus depends on your selected mode and edit type.
Adjust Settings
Configure settings like Quality, Size, Background, and Moderation.
Generate Image(s)
Click the “Generate” button.
Review and Refine
Examine the generated image(s). Iterate by adjusting prompts, settings, or inputs if needed.
Input Parameters and Options
GPT-4o Image offers a range of input parameters, varying based on the selected Mode and Edit Type.Common Parameters (Available across multiple Modes/Edit Types):
Your text description guiding the image creation or modification. (Max 1000 characters).
Controls the level of detail and generation time.
- Possible enum values:
low,medium,high,auto.
The desired output resolution and aspect ratio.
- Possible enum values:
1024x1024,1024x1536,1536x1024,auto.
Choose the background type for the generated image.
- Possible enum values:
opaque,transparent. (Note: Transparency availability depends on Output Format).
Adjusts content filtering sensitivity.
- Possible enum values:
auto(Standard),low(Less Restrictive).
Mode/Edit Type Specific Parameters:
Generate Mode
Generate Mode
Edit Mode (Draw Mask, Maskless Editing)
Edit Mode (Draw Mask, Maskless Editing)
The base image file to be edited. (Max 10MB).
Edit Mode (Draw Mask)
Edit Mode (Draw Mask)
The mask image (Base64 string or URL) defining the area to be edited. Created using the built-in mask editor.
Edit Mode (Reference Images)
Edit Mode (Reference Images)
Up to 4 image files to use as visual references. (Max 1MB each recommended).
Tips for Best Results
Choose the Right Mode & Edit Type
Select the mode and edit type that precisely match your desired outcome (generation, specific edit type, or customization method).
Master Prompting for Your Task
Tailor your prompt content and detail based on the selected mode and tool’s requirements. Be specific and use natural language.
Use High-Quality Input Images
For Edit mode, start with clear, high-resolution images for the best results.
Iterate and Refine
Use multi-turn conversations to adjust outputs until you achieve the desired result.