Image

Use cue image to generate a picture from a prompt, start from a reference image, or edit one you already have.

The Cue backend is the default provider, so it uses your Cue account and credits with no vendor key of your own.

Core Commands

cue image -h
cue image "a cute cat"
cue image "app icon" -o /tmp/icon.png --aspect-ratio 1:1
cue image "make the mug blue" --image mug.png --model <edit-model> -o blue.png
cue image "same scene at night" -c
cue image "a cute cat" --json

Which Models Are Available

Availability is per deployment, so any list written on a page rots. Ask the backend:

cue image --list-models
cue image --list-models --json

That prints the model IDs the current backend serves, their providers, credit costs, and which ones support editing and image-to-image. Today that spans Gemini, OpenAI, SeedDream, xAI, Qwen, and Cue's own image models.

Pick one with --model:

cue image "a cat" --model gpt-image-1

Editing and References

--image (or -i) supplies the input image — a reference to carry a subject or a look, or the picture you want changed. Editing needs a model that supports it; --list-models marks those.

cue image "make the mug blue" --image mug.png --model <edit-model> -o blue.png
cue image "restyle this" --provider seedream --image photo.jpg

Iterating on the Last Result

Each cue image call is fresh by default: no prior conversation context is sent, so earlier generations cannot bleed into a new prompt. Pass -c / --continue to reuse the profile's image conversation as context, which lets a follow-up prompt build on the previous result.

cue image "a lighthouse at dusk"
cue image "same scene at night" -c
cue image "a lighthouse at dusk" -n   # fresh generation; -n wins over -c

Shape and Size

cue image "a cat" --aspect-ratio 16:9
cue image "a cat" --aspect-ratio portrait --resolution 2K

Aspect ratios: 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9. Aliases: square (1:1), portrait (9:16), landscape (16:9).

Resolution accepts 1K, 2K, and 4K where the model supports it. --size passes a generation output size the selected model accepts.

Options

Flag	Description
`--provider <provider>`	`cue` (default), `gemini`, or `seedream`
`--model <model>`	Model to use — run `cue image --list-models` for the live list
`--aspect-ratio <ratio>`	Output shape
`--resolution <res>`	`1K`, `2K`, `4K` when supported
`--image, -i <path>`	Input image to edit or use as a reference
`--size <size>`	Output size accepted by the selected model
`-o, --output <path>`	Output file path
`-c, --continue`	Continue the profile's image conversation
`-n, --force-new`	Fresh generation — the default, and it wins over `-c`
`--list-models`	Print the models this Cue backend advertises and exit
`--json`	Machine-readable output

Direct Providers

--provider can bypass the Cue backend and call a vendor with your own key. This is the escape hatch for someone already paying that vendor, not the normal path.

Provider	What it uses
`cue` (default)	Cue backend — your Cue auth and credits
`gemini`	Direct Gemini API — your own `GEMINI_API_KEY`
`seedream`	Direct ByteDance SeedDream API — your own `ARK_API_KEY`

cue image "a cat" --provider gemini
cue image "a cat" --provider seedream                 # SeedDream v5 (default)
cue image "a cat" --provider seedream --model v3      # SeedDream v3 (cheapest)

Direct Gemini models are gemini-3.1-flash-image-preview (default) and gemini-3-pro-image-preview. Direct SeedDream models are v5 (default), v45, v4, v3, and edit (SeedEdit 3.0, dedicated image-to-image).

Set the keys once:

cue skill env set GEMINI_API_KEY <your-key>
cue skill env set ARK_API_KEY <your-key>
cue skill env list                             # verify configured keys