Generate an image from text
Describe the subject, setting, composition, lighting, and style to create a new image without a source file.
Browse by modality
Image
Generate and edit still images.
What you can do
Image models can turn a description into a new visual, revise an existing image, or combine several references into one result. The right choice depends on whether you value speed, exact editing, typography, reference consistency, or final resolution.
Edit with references
Upload one or more images and explain what should be retained, replaced, restyled, or combined.
Make a local correction
Use a mask in supported models to replace an object, repair a selected area, or change part of the composition.
Prepare production variants
Explore alternative compositions, aspect ratios, and resolutions before selecting a result for a campaign, product page, or publication.
Inputs
- A text prompt describing the desired result
- One or more source or reference images where supported
- An optional mask for models with local editing
- Aspect ratio, resolution, and output controls available for the selected model
Outputs
- A newly generated image or an edited version of the source
- One or more variants, depending on the workflow
- Model-specific output resolution, from fast drafts to high-resolution results
How to choose a model
Start with the kind of control you need. Open a family page to compare its versions, parameters, reviewed examples, and current availability.
GPT Image
Choose GPT Image 2 for detailed instruction following, multi-image editing, and inpainting with a mask.
View model details →
Gemini Image
Use Gemini image models for complex prompts, typography, fast iterations, and reference-based editing across several quality tiers.
View model details →
FLUX.2
FLUX.2 offers a range from quick drafts to production-oriented generation with fine controls and multiple references.
View model details →
Seedream 5
Seedream is useful when you need high-resolution output, editing variants, or a workflow built around several reference images.
View model details →
Reviewed examples
These outputs were reviewed for the public catalog. They show actual model behavior, not stock illustrations.
GPT Image 2
View model details →
FLUX.2 [klein] 4B
View model details →
Limitations and trade-offs
- A model may reinterpret small details, faces, logos, or text even when the prompt asks to preserve them.
- Masks and multiple references are not available in every version; check the selected model's inputs before starting.
- Higher resolution and quality settings usually take longer and cost more than draft generation.
- Review important claims, labels, product details, and brand elements before using an output in production.