Image Generation

Rayu-CLI ships a built-in GenerateImage tool that lets the agent create and edit images from a text prompt, using either NVIDIA's free hosted image models or Google Imagen 4 on Vertex AI. The agent uses it automatically when it needs an image (for example, generating assets for a frontend it is building) or when you ask for one.

Requirements

Either backend enables the tool:

NVIDIA — reuses the nvidia API key from ~/.rayu/providers.json, or the NVIDIA_API_KEY environment variable. Run /connect and pick NVIDIA, or set NVIDIA_API_KEY.
Google Vertex AI (Imagen 4) — uses a configured Gemini / Vertex AI provider (OAuth / ADC). Run /connect → Google Gemini — Vertex AI, or have Application Default Credentials + GOOGLE_CLOUD_PROJECT set. See Providers.

The tool is hidden when neither backend is configured. When both are available, selecting an imagen-* model routes to Vertex; otherwise NVIDIA is used (Vertex is used automatically when it is the only configured backend).

What it does

Calls the NVIDIA genai endpoint (https://ai.api.nvidia.com/v1/genai/<model>).
Saves the PNG to disk (default ./generated-image-<timestamp>.png, always inside the working directory) so generated assets can be referenced from code.
Returns the image inline so the model can see the result.
Displays it in your terminal: native inline image on iTerm2/WezTerm, or truecolor ANSI half-blocks on any 24-bit/256-color terminal (xterm-256color, Kitty, Ghostty, most Linux terminals). Falls back to printing the path.

Parameters

Param	Description
`prompt` (required)	Text description of the image, or the edit to make.
`output_path`	Where to save the PNG (inside the working directory).
`model`	Image model id (see below).
`width` / `height`	Dimensions (FLUX models).
`aspect_ratio`	e.g. `1:1`, `16:9` (Stable Diffusion models).
`steps`, `cfg_scale`, `seed`, `negative_prompt`	Sampling controls.
`input_image`	Path to an existing image to edit (routes to an editing model).

Models

The model list is not built into the CLI. It comes from the Rayu media-model catalog at runtime (GET /v1/models?media=image on the gateway, backed by the admin-managed media_models table), so a model added in the dashboard shows up in /model_image_generation without a CLI update. Run /model_image_generation to see exactly what your account can use.

When you are signed out (or USE_RAYU_OAUTH=false), the CLI falls back to a built-in list so BYOK image generation keeps working offline with your own key.

The shipped catalog defaults are:

Model id	Use
`black-forest-labs/flux.1-schnell`	Default — fast text→image
`black-forest-labs/flux.1-dev`	Higher-quality text→image
`stabilityai/stable-diffusion-3.5-large`	High-quality, `aspect_ratio`/`negative_prompt`
`black-forest-labs/flux.1-kontext-dev`	Image editing (used automatically with `input_image`)

Vertex AI (Imagen)

Available when a Gemini / Vertex AI provider is configured.

Model id	Use
`imagen-4.0-generate-001`	Default (Vertex) — text→image
`imagen-4.0-fast-generate-001`	Faster, lower-cost text→image
`imagen-4.0-ultra-generate-001`	Highest-quality text→image
`imagen-3.0-capability-001`	Image editing (used automatically with `input_image` on Vertex)

Video generation (`GenerateVideo`)

The companion GenerateVideo tool generates short videos from a text prompt. It is enabled by NVIDIA/fal.ai keys or by a Gemini / Vertex AI provider. On Vertex it uses Veo 3.1 (veo-3.1-generate-001, veo-3.1-fast-generate-001) via the long-running prediction API (Rayu polls until the video is ready, then saves the MP4 inside the working directory).