Cremi CLI

Beta

Run Cremi's generation tools straight from your terminal.

Prefer the terminal, or want to keep your coding agent's context lean? The CLI wraps the same tools in a scriptable interface with JSON output and meaningful exit codes.

1.Get your access token

The CLI signs in with the same personal access token as MCP, so if you've already set one up you can skip straight to the install below.

How to create a token →

2.Install and authenticate

Install

npm install -g @cremi_ai/cli

Published on npm as @cremi_ai/cli. Requires Node.js 18 or newer.

Authenticate

cremi login

Paste your access token when prompted — it's validated and stored in ~/.config/cremi/config.json.

3.Check it works

Before spending credits, confirm the token is valid and see which tools your account can reach.

Examples

# confirm the token works and see which tools you have
cremi doctor
cremi tools

4.Generate an image

`cremi image` takes the prompt as its only required argument. Everything else has a default, so start simple and add options as you need them.

cremi image

# simplest form — prompt only, everything else defaults
cremi image "neon koi fish swimming through clouds"

# choose a model, shape and save location
cremi image "dreamy lofi album cover" \
  --model nano-banana-pro \
  --aspect-ratio 1:1 \
  --resolution 2K \
  --output covers/lofi.png

# edit or reference existing images (nano-banana family, flux-pro-ultra)
cremi image "same character, now in the rain" \
  --model nano-banana \
  --reference https://example.com/character.png

Options

--model: Which model renders the image. Defaults to imagen-4-fast, the cheapest.
--aspect-ratio: Shape of the output: 16:9, 3:2, 4:3, 1:1, 3:4, 2:3 or 9:16.
--resolution: 1K, 2K or 4K. Only the nano-banana family and gpt-image-2 honour this; other models ignore it.
--output: Where to save the file. Pass a directory to keep the generated name, or a filename to set your own.
--reference: One or more image URLs to work from, for edits and style references. Repeat the flag for several.

Without --output the command prints the image URL and leaves nothing on disk. With it, the CLI downloads the full-resolution original rather than the smaller preview shown in chat clients.

Model	Notes
imagen-4-fast · imagen-4-standard · imagen-4-ultra	Google Imagen 4 — fast, standard and ultra quality tiers
flux-pro · flux-pro-ultra	FLUX 1.1 Pro and Pro Ultra
openai · gpt-image-2	OpenAI gpt-image-1, and GPT Image 2 — 1K / 2K / 4K resolution, up to 16 reference images
nano-banana · nano-banana-pro · nano-banana-2	Nano Banana family — editing and reference images, 1K / 2K / 4K resolution

5.Generate a video

`cremi video` takes the prompt as its only required argument and passes your flags to generate_video. Every media flag accepts a URL or a local file — local files are uploaded to your Assets first. Renders take minutes, so the command waits for a while and then hands back a task id you can collect later.

cremi video

# text to video — returns a task id, then polls until it is ready
cremi video "a red apple slowly rotating on a wooden table" \
  --model grok-imagine --duration 5 --aspect-ratio 16:9

# animate a still image (required for wan-2.6-flash) — a URL or a local file
cremi video "gentle slow zoom" --model wan-2.6-flash --duration 5 \
  --image ./first-frame.jpg

# native lip-sync: the singer as a reference image, the vocal clip as reference audio
# (local files are uploaded to your Assets first; --quote prices it without charging)
cremi video "she sings the line to the camera, close-up" \
  --model minimax-h3 --duration 6 \
  --ref-image ./singer.png --audio ./vocal-6s.mp3 --quote

# reference-to-video on Seedance: images + a motion reference clip
cremi video "the dancer repeats this move on a rooftop" \
  --model seedance-2.0-fast --resolution 480p --duration 5 \
  --ref-image ./dancer.png --video ./move-5s.mp4

# any parameter is still reachable through the generic form
cremi run generate_video prompt="a paper boat drifting down a rainy street" \
  model=seedance-2.0-fast resolution=480p duration=4

# collect a render that outlived the call, and see live credit costs
cremi run get_video_task_status task_id=<TASK_ID>
cremi run list_video_models

Parameters

<prompt>: What to generate. Required.
--model: Which model renders the video. See the table below.
--duration: Length in seconds. Each model accepts its own set of values — anything else is rejected before you are charged.
--resolution: Output resolution, for models that offer more than one (see the table below). It changes the price, and anything the model does not list is rejected before you are charged.
--aspect-ratio: 16:9, 9:16 or 1:1.
--image: First frame, as a URL or local image. On reference models it is the first reference image instead. Required by image-to-video models.
--end-image: Last frame the clip must land on, for seamless chaining. Only models that support it accept it.
--ref-image: Reference images for subject and look, not frames. Repeat the flag or list several.
--audio: Reference audio (.mp3 or .wav, 2–15s) that drives timing — and native lip-sync on minimax-h3.
--video: Reference video (.mp4 or .mov, 2–15s) for motion. Billed for its length as well as the output.
--quote: Print the exact credit price for these inputs and stop. Nothing is charged.

If the render outlives the call you get a task id instead of a URL. Credits are charged when the job is accepted and returned automatically if it fails.

Model	Notes
grok-imagine	Cheapest option — 720p, 1–15s
kling-2.1-standard · kling-2.5-turbo-pro	Kling 2.1 Standard and 2.5 Turbo Pro — 5s or 10s
veo-3.1-fast-silent · veo-3.1-fast · veo-3.1	Google Veo 3.1 — silent, fast and full quality tiers; the non-silent tiers add a native audio track. 4 / 6 / 8s
seedance-2.0 · seedance-2.0-fast · seedance-2.0-mini	ByteDance Seedance 2.0 family — 2.0 at 480p / 720p / 1080p, Fast and Mini at 480p / 720p, 4–15s. Choose with resolution; the default is 720p. Takes up to 9 reference images, 3 audio and 3 video clips; references replace the first/last frame, and audio needs an image or video beside it
seedance-2.5	ByteDance Seedance 2.5 — 480p or 720p, 4–15s. image_url and up to 9 reference images steer subject and look rather than setting the first frame. Also takes up to 3 audio and 3 video references
minimax-h3	MiniMax H3 — 720p or 2K, 4–15s, native audio. Needs image_url or up to 5 reference images, used for subject and look rather than the first frame. Reference audio (native lip-sync) or video moves the render to MiniMax's own API, which is priced higher
minimax-h3-max	MiniMax H3 Max — the faster H3: 480p or 720p, 5–15s, native audio. Up to 2 reference images (image_url counts as one), 3 audio and 3 video references
wan-2.6-flash	Wan 2.6 Flash — image-to-video only, so image_url is required. 5 / 8 / 10 / 15s
wan-3	Wan 3.0 — 480p / 720p / 1080p, 4–15s. Up to 10 reference images, 5 audio and 5 video references; images are references, not the first frame. With reference audio the clip keeps a soundtrack
pixverse-v6 · pixverse-c1	PixVerse V6 and C1 (C1 is the cinematic tier) — 720p or 1080p, 4–15s, no audio track. Needs image_url or up to 7 reference images, which the video is built from; cite them in the prompt as @Image1, @Image2 …

6.Use it in scripts

Every command takes --json, and `cremi run` can call any tool by name — including ones added after your CLI version, so you do not need to upgrade to reach them.

Use it in scripts

# --json prints a machine-readable result instead of prose
cremi image "album art" --json | jq -r .url

# `run` calls any tool by name, so new tools work without a CLI upgrade
cremi run generate_image prompt="misty forest" model=imagen-4-ultra --json

The CLI talks to the same MCP server, so every new tool appears automatically in cremi tools and cremi run.

Credit costs are not listed here because they come from the live pricing table. Run `cremi run list_video_models` for current video prices, or check Usage in your account.