Generating images

Describe the image's purpose

Ask for images in the project chat. Explain where the result will appear, its composition, colors, visual style, and whether text must be included. Attach a relevant reference image when it helps.

For example:

Create one square illustration for the empty state on the bookings page. Use our muted blue palette, leave space around the subject, and do not include text. Attach it and add it to that page.

Amazi uses the image model and options available to the current task. The available model is listed in the current catalog in AI Gateway and applies to generated assets and design references. Generation may be unavailable if the required capability or model is not enabled.

Design references for a new application

Attach an image or provide a website link when you have a visual reference. Explain what to preserve and what should change for your product. Amazi's design workflow uses that reference to describe screen structure, wording, style and imagery before implementation.

When starting a new visual application without a reference, Amazi prepares a generated design image first. The image request asks for a representative screen based on selected original messages and your answers from the current chat. Extractable text from their attachments provides additional context; reference images are selected separately. Scanned pages are not recognized automatically. One screen can establish the look of several pages; it does not reduce the pages or features you requested. Generation availability and credit charges still apply.

For new visual work and routine updates, Amazi first uses suitable images you provide and preserves suitable existing project images. It generates missing photos and illustrations needed for your request automatically, using low or medium quality by default. You do not need a separate request to generate each necessary asset. Say if you want to prohibit generation or limit its cost. Icons come from an icon library; Amazi does not hand-draw SVG assets, and can reuse SVG files you supply.

Selected image inputs

For image generation, Amazi can use selected conversation messages, your answered questions and extractable text from their attachments alongside the image request. Only selected reference images are sent as images; unrelated chat attachments are not automatically included. PNG, JPEG and WebP can be used directly from chat attachments or workspace files. PDF and office documents supply extracted text, not rendered pages; scanned text is not recognized.

Quantity is the number of output files

One generation request can produce between 1 and 4 variations of the same description, depending on the enabled model. For different subjects, scenes, or styles, Amazi generates each image separately with its own description. For example, a honey cake, an anthill cake, and a chocolate cake need three separate generations. Shared lighting, background, and photographic style can keep the set consistent. Increasing the number of outputs does not assign different subjects to individual files.

If you want several variations of one image, say so: those can be generated together, up to four at a time.

Amazi can run up to four independent generations at once within a task, each producing up to four variations. It can continue other work while images are being prepared, then collect and inspect the results before completing the task. If one generation fails, successful images from the others remain available.

“One image with three objects” and “three separate images” are different requests. Say which you need. List the composition of each image and specify separate files or a collage. Amazi should inspect the results before reporting the complete set as ready.

Quality and cost

Explain the display size and intended use. Amazi supports low, medium, and high quality. It defaults to low for simple images and medium for more detailed assets. Ask for high quality when you need finer detail; higher quality and larger images can cost more. Automatic quality is unavailable.

You can request an exact image size, including square, landscape, portrait, 2K, and 4K images. Custom dimensions must be multiples of 16, neither side may exceed 3840 pixels, and the longer side must be at most three times the shorter. Total area must be between 655,360 and 8,294,400 pixels. For example, 1536×864, 2048×2048, 3840×2160, and 2160×3840 are supported. Sizes above 2560×1440 are experimental. You can also let the model choose the size automatically.

Generation is metered work. The credit total shown for a completed task can also include reasoning, coding, and other operations; it is not an image-only invoice. Actual availability and billing depend on the configured service. See Plans and credits.

Download, reuse, or integrate

Open the work summary to follow each generation: an animated placeholder stays visible while images are being prepared and loaded, then the generated images appear in its place. Previews are at most 160 pixels high and preserve the image proportions. These previews, including generated design references, are visible to people with access to the project and remain in the conversation after a reload. Select an image to open it in the preview dialog, where you can view it at a larger size or download it. Failed or interrupted generations show their status instead of continuing to load.

Images requested as downloadable deliverables can also appear as reply attachments. If the image already exists, ask Amazi to reuse it. Request a new generation only when the result needs to change.

An attached image is not automatically part of the application. If you asked for integration, verify the target page in Preview, including its cropping and mobile layout. Downloading the result and placing it into source are separate outcomes.

For visual changes to an existing layout, a screenshot and a precise description can be more useful than generating a new asset. See Files and attachments and Testing and verification.