Generate video from GPT-6 Astra

That is the entire loop, unedited, from a real session: one sentence in, a clip out, one line on the bill. The clip above is the generated video, unedited. The work goes through Actionway's video generation capability, and the Skill installed with the CLI is what teaches the agent to plan, price, run, and deliver it.

What happened in between

OpenAI released GPT-6 Astra on September 3, 2026, positioning it around computer use and long-running agent tasks — a framing TechCrunch's launch coverage shares. Since Codex CLI 0.153.4 it is the bundled default model, so a fresh Codex session already runs on it. What it still lacks is a way to render pixels, and that is where the Skill comes in.

Before answering, the agent checked the capability index it keeps locally, so the first reply already carried the plan: which capability, which models, what each costs per second. It presents those options neutrally — no default, no recommendation; say "you pick" and it takes the best-quality tier. Picking one is the approval: the call runs at the shown price, and Actionway checks the wallet on its side. Only runs the platform flags for extra confirmation pause for another question first. Video runs as a job, so the agent waits on it rather than polling by hand, then downloads the finished clip into outputs/ with a descriptive filename. If the wallet cannot cover a run, the agent stops before anything is charged and gives you a top-up link with the amount prefilled.

Pick a model, pay per run

Video is priced per second of output, and the resolution tier moves the price. The Seedance family spans the range: Seedance 2.0 Mini (from $0.07 per second (480p)) for drafts, Seedance 2.0 (from $0.14 per second (480p)) up to 4K, and Seedance 1.5 (from $0.12 per second (720p)) in between. Kling — Kling 3 from $0.20 per second (720p) — gives a second look for the same brief. The prices are rendered from the same registry the models library shows, so they match what your agent quotes.

A five-second draft on the mini tier is cheap enough to iterate on the prompt; the premium run is for the frame you will actually ship. The same capability also accepts a still image or a reference clip as the starting point, which is how a product photo turns into a short e-commerce remix. Every attempt is one catalog-priced run, and there is no subscription — a balance you top up once covers every capability in the catalog.

Writing a prompt a video model can follow

Video prompts carry one thing image prompts do not: motion. The case prompt names a subject, an action, a camera move, and a light, in that order, and that is about as much as a five-second clip can hold. Describe one movement, not a sequence — "the camera slowly pushes in" works, "the camera pushes in, then pans left, then the robot turns around" produces a clip that does none of them well. If the subject matters more than the motion, start from a still image instead: the same capability accepts a photo as the first frame, and the prompt then only has to say what moves.

Check every draft against the brief before spending on a higher tier. Video models invent details to fill a frame — lettering on a wall, a logo on a product, a second character at the edge — and the cheapest fix is a phrase in the prompt ("no text, no logos") plus one more draft on the mini tier at $0.07 per second (480p). Only when the draft does what the sentence says is it worth running the same sentence through Seedance 2.0 at 1080p or 4K, or through Kling 3 for a different look.

Set it up once

npm install --global @actionway/cli@latest
actionway init
mkdir -p "$HOME/.codex/skills"
cp -R "$HOME/.actionway/skill/actionway" "$HOME/.codex/skills/actionway"

You need Node.js 20 or newer and an Actionway account; after sign-in the agent reports your wallet balance and, if it is empty, links you straight to top-up. Prefer no local CLI? Codex can also load Actionway as a plugin — codex plugin marketplace add PawLogic/actionway-plugins then codex plugin add actionway@pawlogic — which runs the same catalog over Remote MCP. The get started page lists every route per client side by side.