Image Generation

Generate images from text prompts, reference images, or natural-language editing instructions.

Generate product photos, marketing visuals, and design concepts by describing what you want. Attach reference images for more control. All results are saved to your Pixa workspace.

Text-to-image

Use the generate_media tool with a prompt and model ID. Call models (action: list) first to find available model IDs:

Generate a studio product photo of a water bottle on a marble countertop.

The agent calls models to find a model, then generate_media with the chosen model ID and prompt.

Reference-based generation

Pass image URLs or asset IDs in the attachments parameter of generate_media. The model uses the attachment as a visual reference while composing the new scene:

Generate a lifestyle scene of this product on a kitchen counter with morning light.

Generation parameters

The generate_media tool accepts:

Parameter	Description	Example
prompt	The text description (required)	"product photo of sneakers on marble"
model	Model ID for generation (required)	nano-banana-2
aspect_ratio	Output aspect ratio	1:1, 16:9, 4:3, 9:16
media_type	image or video	image
output_format	File format	png, jpg, webp
num_variations	Number of variations to produce	1-4
attachments	Reference image URLs or asset IDs	—
brand_library_id	Brand library for brand-consistent output	lib_abc123

Browsing models

The models tool supports list, search, get, and recommend actions. Use recommend with a natural language description of your use case to get ranked suggestions:

Find a good model for generating product lifestyle photos.

Batch generation

Generate multiple variations in a single call by setting num_variations:

Generate 4 lifestyle photos of this watch on a wooden desk.

To produce the same concept across different aspect ratios, ask for separate runs:

Generate this product shot in 1:1 for Instagram, 16:9 for the website banner, and 9:16 for stories.

Prompt engineering tips

Write prompts as if briefing a photographer: subject first, then environment, lighting, and mood.

Technique	Example
Be specific about the subject	”white ceramic mug” instead of “a mug”
Describe the environment	”on a rustic wooden table, blurred garden background”
Specify lighting	”soft natural light from the left, gentle shadows”
Set the mood or style	”minimal, editorial, high-end product photography”
Mention camera details	”shot at eye level, shallow depth of field”

When passing a brand_library_id, the model pulls brand colors, fonts, and style references automatically. Combine this with a descriptive prompt for brand-consistent output.

Related

• Image Generation (User Guide) — in-app generation workflow
• MCP Server — full MCP tool reference