The Flova Skill: One Token, and a Fork in the Road

What Flova Does

You describe a video in plain language. Flova handles the scriptwriting, storyboard design, image, video and audio generation, timeline assembly and export, then hands back a project link and the finished file. You do not have to pick a model to get started — Flova selects for you — and the skill names Sora, Veo, Kling, Seedance, MiniMax, HaiLuo, Jimeng, Vidu, Grok and Wan among the tools it covers. If you do want a specific one, asking for it is just another message.

Architecturally it is the same bargain as the LibTV skill: the creative decisions happen on the vendor's side, and your agent's job is to relay. Flova's own skill calls this a "creative passthrough" — every request, whether that is a new script, a model switch or a storyboard rebuild, goes over as a natural-language message rather than as a parameter your agent has to construct.

The Two Install Routes

Flova ships two, and they work quite differently.

• The ClawHub package — openclaw skills install @flova/flova-video-generator. Published by @flova themselves, MIT-0. It is instructions plus curl: no binary, no package, nothing to compile. This is the one we host.
• The Flova CLI — the route advertised on their Agent CLI page. You hand your agent a setup prompt; it fetches an install script from their CDN and runs it through a shell, which copies a flova binary into a bin directory and a skill into ~/.flova. The setup prompt then asks your agent to link that skill into wherever your client indexes skills.

Our marketplace policy is to reject anything that pipes a freshly fetched script into a shell, because what runs is whatever that URL serves at that moment, with your permissions and no review step. That policy is why we host the installer-free package rather than the CLI. It is a rule about how software arrives, not a judgement about Flova — plenty of reputable tools ship this way. If you install by hand and would rather not run an installer, the ClawHub package is the one to reach for.

Getting the Token

Flova authenticates with a single bearer token in FLOVA_API_TOKEN. Get one from the Agent CLI page, which carries a "sign in and get a key" link for exactly this.

Worth knowing before you go looking: the skill still hardcodes an older token URL, flova.ai/openclaw/?action=token, and that address returns a 404 as of August 2026. If your agent quotes it at you, ignore it and use the Agent CLI page instead.

export FLOVA_API_TOKEN="your-token"

That is the entire dependency list, alongside curl. Both our OpenClaw and Hermes images ship curl, so the skill runs on either. Generation is billed by Flova against your own account, not by us.

How a Job Runs

• Your agent creates a project, or reuses the one from this session.
• Your wording goes over as a chat message, with any uploaded reference files attached.
• It polls every 3 seconds while relaying new messages and asset URLs to you, and gives up after 20 minutes rather than spinning forever.
• When the timeline is composited, it exports, polls the export, and returns the finished video plus the project link.

At a minute or under, the skill has a fully-managed mode: it runs that loop on its own and auto-confirms the low-stakes questions like style, music and layout, surfacing only decisions with real consequences — payment, or a content warning. Above that length it stays in the normal flow, where every question Flova raises gets passed to you and waits for your answer.

What We Changed Before Hosting It

Our copy is not byte-identical to Flova's published version, and you should know what differs. We read every line before listing it and made three changes, none of which remove a capability:

None of this suggests Flova is acting in bad faith; a version check and a tips feed are ordinary product decisions. But a skill runs inside your agent with your token, and "fetch remote text every turn and put it in front of the user" is a channel worth closing when the feature it powers is cosmetic.

Flova or LibTV?

We host both, and they overlap heavily: both are conversational video agents, both do script to storyboard to finished cut, both ask your agent to relay rather than create. Pick on the details. Flova's skill is pure curl with a single token, and it will drive a short piece end to end without asking you anything. LibTV bundles Python scripts and gives you a canvas you can open and edit by hand alongside the agent. Both bill generation to your own account, so trying one costs you a signup rather than a subscription to us.