Generate Image
Generate an image from a text prompt using AI. Spends AI tokens per generation.
Category: AI Menu path: AI > Generate Image
Ports
| Port | Type | Direction | Description |
|---|---|---|---|
prompt_in | string | input | Optional — drive the prompt from a String node instead of typing it |
image_in | imageRgba16f | input | Optional — an upstream image as an image-prompt (img2img). Use a direct Image Source or another Generate Image, and pick an image-capable model |
out | imageRgba16f | output | The generated image (transparent until you generate one) |
Parameters
All parameters are managed by the Generate panel in Properties (not standard rows):
| Param | Type | Default | Description |
|---|---|---|---|
prompt | string | "" | What to generate. Be descriptive. |
model | string | (first available) | Which AI model to use — a small curated list. |
aspectRatio | string | 1:1 | Output shape: 1:1, 16:9, 9:16, 4:3, 3:4. |
resultPath | string | "" | Hidden — set automatically to the saved image under the project's generated/ folder. |
How It Works
When you click Generate, Caddis sends your prompt to the Caddis server, which deducts the model's token cost from your balance and runs the generation on a curated provider model. The resulting image is downloaded into your project's generated/ folder, and the node loads it like an image source — so it composites and caches exactly like any other image. If generation fails, your tokens are automatically refunded.
The token cost for each model is shown on the Generate button. You need a saved project (so generations have a home) and enough tokens. If you're short, the Generate button opens the token top-up dialog.
Usage Examples
Basic: Generate a background
Add a Generate Image node, type a prompt ("misty pine forest at dawn, cinematic"), pick an aspect ratio, and click Generate. Wire its output into a Transform2D → Output to place it on the canvas.
Iterating
Edit the prompt and hit Re-run to produce a new image. Each run saves a fresh file in generated/; the node points at the latest. Import to Assets adds the result to the asset manager; Delete removes it.
img2img (image prompts)
Pick an image-capable model (e.g. Nano Banana 2) and an image_in port appears (text-only models like Flux Schnell show no image port). Wire in one or more images — Image Sources, prior generations, or any rendered node (DrawShape, processed chains; captured via the engine) — and the generation riffs on them (multiple images are blended). Reorder or remove inputs in the Properties list.
Running low on tokens
If your balance is below a generation's cost, the Generate button becomes "Get tokens · N more" and opens the top-up dialog — the server never charges or fails a generation you can't afford.
Graph-driven prompt
Wire a String node into prompt_in to compute the prompt from elsewhere in the graph.
Tips
- Token cost is per generation and is shown on the button — re-running spends again.
- Generations live in your project's
generated/folder; they travel with the project. - The node shows transparent until you generate, so an empty Generate Image won't break a comp.
Related Nodes
- ImageSource — Load a still image from disk (no AI / no tokens)