Image generation
Turn a prompt into a file on disk — and, on a capable terminal, an image drawn inline right after the command.
melious image "a sunset over the Swiss Alps, oil painting style"Files land in the current directory as melious-{timestamp}.{ext}. On a supported terminal (Kitty, iTerm2, Windows Terminal with Sixel) the image also renders inline. Everywhere else, you get the path.
Flags
| Flag | Type | Default | Description |
|---|---|---|---|
--model | string | flux-1-schnell | Image model |
--size | string | 1024x1024 | Dimensions, e.g. 1792x1024 |
--quality | string | standard | standard or hd |
--style | string | vivid | vivid or natural |
-n, --count | int | 1 | How many images to generate |
-o, --output | string | . | Output directory (created if missing) |
--seed | int | — | Seed for reproducibility |
--steps | int | — | Inference steps (model-dependent) |
--guidance | float | — | Guidance scale (model-dependent) |
--negative | string | — | Negative prompt |
Examples
A batch of four:
melious image -n 4 "cyberpunk cityscape at night"Drop renders into a folder:
melious image -o ./renders "abstract geometric patterns"HD with a specific model:
melious image --model flux-1-dev --quality hd "portrait of a cat reading a book"Reproduce a specific result:
melious image --seed 42 "minimalist logo design"Squeeze more control out of the sampler:
melious image --steps 50 --guidance 7.5 --negative "blurry, low quality" "a serene mountain lake"Inline preview
| Terminal | Protocol |
|---|---|
| Kitty | Kitty graphics |
| iTerm2 | iTerm2 inline images |
| Windows Terminal | Sixel (if enabled) |
If we can't detect a supported protocol, we save the file and print the path. That's the one case where the CLI goes quiet — the image still exists, we're not drawing it for you.
Machine-readable output
--json returns metadata (model, dimensions, paths) without the image bytes — useful when you're calling melious image from a script and only want to know where the files ended up:
melious image --json "a red circle"Code review
Run a model over a git diff and get review notes back — bugs, security issues, performance problems, style. Not a replacement for human eyes, but the pass you run before you ask for those eyes, to catch the obvious stuff.
Coding tools
A handoff. We generate the config for a coding tool, set the environment it needs, and start it. You spend no time figuring out where each tool keeps its provider config or which env var it reads for an API key.