Higgsfield CLI generates AI images and video from a terminal

Higgsfield's command-line tool generates AI images and video from a terminal, a script or a coding agent, and the company published a step-by-step guide to the workflow on October 19. The Higgsfield CLI runs the same model catalog as the web app from the same account and the same credit balance, with no browser tab and no API key.

That last detail is the one to settle before installing. Higgsfield bills CLI work to plan credits at standard rates, and the free generations bundled with an Unlimited plan do not follow you into the terminal.

What is Higgsfield CLI?

Higgsfield CLI is an open-source command-line client for Higgsfield, an AI-native creative suite for image and video generation. The code and releases are on GitHub under an MIT license, and the repository was created on April 29, 2026. The command is higgsfield, with higgs as a shorter alias, and the npm package is @higgsfield/cli.

The maker's guide frames the tool with three rules. There is no API key, because sign-in happens through the browser. Every generation is charged to the account's plan credits at standard rates. And web-only access stays on the web: unlimited model access and free generations do not apply to CLI generations.

The CLI reads the same catalog as the web app. The GitHub README says "40+" models, while the guide's own table says "35+ models and tools" covering image, video, 3D and audio, plus analysis such as the Virality Predictor. Higgsfield explains the gap. higgsfield model list, run on October 16, returned 107 entries because the catalog lists each mode and variant as a separate entry, including upscalers and background removers. The named examples run from Seedance 2.5 and Kling v3.0 to Google Veo 3.1, Gemini Omni Flash, Nano Banana Pro, FLUX.2 and Soul V2.

How do you install and sign in?

There are three install routes, and updates go through whichever one you used. The curl installer covers macOS and Linux. Homebrew covers the same two platforms with brew install higgsfield-ai/tap/higgsfield. npm covers any platform, including Windows, with npm install -g @higgsfield/cli.

Sign-in is the part that differs from most command-line tools. You run higgsfield auth login, the terminal opens a browser, and you approve a consent page listing what the CLI will access. The terminal then prints your email, plan name and available credits.

Two requirements come first. You need a paid Higgsfield subscription, and the machine needs macOS, Linux or Windows. The guide's troubleshooting section is worth reading once. In the maker's own test on macOS, the curl installer reported version 1.1.26 while higgsfield version printed an older 1.0.1 left over from a Homebrew install, and which -a higgsfield showed both copies.

How do you check the cost before you generate?

Run higgsfield generate cost with the same model id, parameters and media flags you would pass to generate create. It prints the credit price and creates nothing.

The maker's test gives two figures worth remembering. A 5-second Seedance 2.5 clip at 1080p priced at 80 credits for a full render, and 15 credits with --draft true at the same duration and resolution. The still image in the same test finished in about a minute at 2752 by 1536 and cost 2 credits. Higgsfield verified those credit costs on October 16, 2026 for one account, and warns that prices depend on model, duration and resolution.

Cost estimation also covers workflows, through higgsfield generate cost workflow. Two workflows, voice-change and dubbing, do not support it.

How do you use it in scripts and coding agents?

Two features make the CLI run without a person at the keyboard. --json works on every command and returns raw records. Local file paths passed to --image, --video or --audio are uploaded automatically, which makes --start-image ./first.png a valid way to animate a file on disk.

Without --wait, higgsfield generate create returns at once with a job id. A loop over a prompt file therefore starts one job per prompt line, which is the batch workflow the guide documents. In the maker's test, a call without --wait printed a one-element list holding the job id while the generation kept running on the server.

Coding agents use the same tool. Higgsfield lists Claude Code, Codex, Cursor, OpenClaw and Hermes Agent as supported clients, and the agent can run the setup itself: install the package, start auth login for you to approve in the browser, then add the companion skills with npx skills add higgsfield-ai/skills. Separately, computingforgeeks notes that Higgsfield also names NemoClaw among the supported agents.

Chat agents take a second route. Higgsfield runs an MCP server at mcp.higgsfield.ai for Claude, ChatGPT and any other MCP client. It shares one account and one credit balance with the CLI, so switching between the two changes the interface rather than the setup.

What does the credit rule change?

Every generation through the CLI or the MCP server spends credits, whatever the plan. Higgsfield's rule is that unlimited access and free generations "apply only on higgsfield.ai," and the guide repeats it for the CLI. The adtestbench guide, which checked Higgsfield's pages on October 21, 2026, says a model that costs nothing on the website still costs its standard rate from a connected agent, and puts the cheapest paid plan at $15 a month. RuntimeWire reported the same standard-rate billing when it covered the October 19 guide.

The separate Higgsfield API is a different product, with its own key from open.higgsfield.ai and a prepaid balance in US dollars. It does not draw on plan credits.

One timing note for long jobs. The default wait is 10 minutes, raised with --wait-timeout. If a generation runs longer, the command can time out while the job keeps running on the server, which is what happened to an earlier text-only video in the maker's test. higgsfield generate get <job_id> still reports its status.

What is still unverified

Higgsfield's figures come from one account on one Mac on October 16, 2026, and the guide says so. Another account's credit prices, model ids and parameters can differ, which is why model list, model get and generate cost exist. The 107-entry catalog count moves with every release. And no independent test of CLI output quality has been published, so any judgment of the clips themselves waits for one.

Install with curl -fsSL https://raw.githubusercontent.com/higgsfield-ai/cli/main/install.sh | sh, or brew install higgsfield-ai/tap/higgsfield, or npm install -g @higgsfield/cli, then run higgsfield auth login and approve the browser prompt.