How to Make Videos With Codex Using Muse Spark Open (2026)

UPDATED: 2026-10-19

To make videos with Codex using Muse Spark Open, install the Pexo skill into OpenAI Codex, point Codex at Muse Spark Open as its backend model, then describe the video you want in plain language. Codex plans and dispatches the job, and the Pexo skill generates and edits a finished clip. Muse Spark Open reasons, Codex orchestrates, and Pexo makes the video.

There is no "make video" button inside Codex, and Muse Spark Open does not render footage either. Keeping that split straight is the whole trick: Muse Spark Open is Meta's reasoning model, OpenAI Codex is the coding agent that runs commands, and Pexo is the installable skill that adds video generation. Muse Spark Open shipped on October 15, 2026 as Meta's open-weight, multimodal mixture-of-experts model — 816 billion total parameters with about 42 billion active per token, a roughly one-million-token context, and a design aimed at long-horizon agentic work. It speaks the OpenAI API format through hosted providers, and that compatibility is exactly why it drops into Codex as a backend. This guide walks the five steps end to end, from configuration through export.

How to make videos with Codex using Muse Spark Open: the quick version

1. Install OpenAI Codex — install the CLI and run codex once so it writes ~/.codex/config.toml.
2. Point Codex at Muse Spark Open — add a provider block with your host's base URL, key, and OpenRouter model meta/muse-spark-open.
3. Install the Pexo skill — run npx skills add https://github.com/pexoai/pexo-skills --skill pexo-agent.
4. Describe the video — type a plain-language brief; Codex plans on Muse Spark Open and the Pexo skill renders.
5. Iterate and export — request changes, pick 16:9, 9:16, or 1:1, and download the finished video.

The five steps in full

Step 1: Install and launch OpenAI Codex once

Install OpenAI Codex and run it a single time so it creates its config file. Install the CLI with npm install -g @openai/codex (or use the official installer for your platform), then run codex in a terminal once. That first launch writes ~/.codex/config.toml on macOS and Linux, or %USERPROFILE%\.codex\config.toml on Windows, which is the file you edit in the next step. Keep in mind that Codex is a coding agent that writes and runs commands, not a video engine, so this install gives you the orchestrator, not the generator.

Step 2: Point Codex at Muse Spark Open as the backend model

Point Codex at Muse Spark Open by adding a provider block to ~/.codex/config.toml. Because the weights are open, several hosts serve the model behind an OpenAI-compatible endpoint; this example uses OpenRouter. Create a key in your provider's console and set it in your shell with export OPENROUTER_API_KEY=... (setx OPENROUTER_API_KEY ... on Windows), then add this block:

model = "meta/muse-spark-open"
model_provider = "openrouter"

[model_providers.openrouter]
name = "OpenRouter"
base_url = "https://openrouter.ai/api/v1"
env_key = "OPENROUTER_API_KEY"

Start a new Codex session with this provider. In a disposable folder, ask Codex to create route-check.txt and read it back. Verify the file and check your OpenRouter activity for meta/muse-spark-open before starting a video project.

Step 3: Install the Pexo skill into Codex

Install the Pexo skill into Codex so the agent can actually generate video. Run npx skills add https://github.com/pexoai/pexo-skills --skill pexo-agent, then add your Pexo credentials when the skill prompts for them. This is the layer that does the rendering: Codex and Muse Spark Open handle the reasoning and command execution, and the Pexo skill turns the plan into finished footage. Pexo is an installable skill you add to Codex, not a feature built into Codex or into Muse Spark Open, so this step is what unlocks video output in the first place.

Step 4: Describe the video you want in plain language

Describe the video you want to Codex in plain, conversational language. Type a brief such as: "Make a 20-second vertical promo for a cold-brew coffee brand, upbeat, with a voiceover and background music." Codex, reasoning on Muse Spark Open, plans the job and dispatches it to the Pexo skill. Pexo auto-routes each shot across models like Seedance 2.0, Kling AI, and more, layers in voiceover, music, and Foley sound effects, and assembles a coherent cut. You do not write model prompts or open an editing timeline — you describe the outcome and Pexo builds it.

Step 5: Iterate on the result, then export

Review the first cut, ask Codex for changes in the same conversation, then export. Tell it what to fix in plain language — "make it shorter", "swap the opening shot for the product close-up", "add captions" — and the Pexo skill re-renders the affected parts. When the cut is right, choose your aspect ratio (16:9 for YouTube, 9:16 for Shorts and TikTok, or 1:1 for feed posts) and download the finished video. Because the whole loop runs inside Codex, you iterate by conversation instead of re-editing in a separate app.

How to get more out of the Pexo skill in Codex

Once the workflow runs, the Pexo skill is where the quality lives, and it repays knowing what it can do beyond a single line. Pexo works from more than text: you can hand Codex an image, a URL, an audio file, a PDF, or a slide deck and ask for a video built from it, which helps when you already have source material and do not want to start from a blank brief. Under the hood it draws on capabilities like text-to-video and script-to-video, so a rough outline or a finished script both become valid starting points.

The multi-model routing is the point. Instead of committing to one generator, Pexo spreads shots across Seedance 2.0, Kling AI, and more, and chooses per shot, so you get breadth without picking models yourself and without switching apps. To see the entry point the skill mirrors, the social media video use case shows the same description box and preset chips you drive through Codex.

Because everything happens inside Codex, you never leave the agent to make the video: no separate account tab, no export-and-reimport shuffle. That single-surface loop — describe in Codex, get a finished video back — is what separates this from wiring a video API into your own script by hand.

Common pitfalls

• Expecting Codex or Muse Spark Open to render the video themselves. They plan and reason; only the Pexo skill generates footage. Muse Spark Open takes text, image, audio, and video in and returns text, so if you skip Step 3 there is no video capability to call and the job stalls — install the Pexo skill and let it handle generation.
• The banner still shows the default model. If codex does not start on Muse Spark Open, you probably put the provider block in a project-level .codex/config.toml, where Codex ignores model_provider for safety. Move the block to ~/.codex/config.toml.
• Codex returns a /responses 404. Check the base URL and the provider documentation for the API route your installed Codex version expects. Resolve the connection before submitting a video job.
• Wrong model ID. Use meta/muse-spark-open with the OpenRouter configuration shown above.
• Hard-coding the key. Do not paste your provider key into config.toml; set env_key to the environment variable name and keep the secret in your shell environment.

Related tutorials

• How to make videos with OpenAI Codex — the same skill running on Codex's default model.
• How to make videos with Codex using Grok 4.6 — the same workflow with another OpenAI-compatible backend that needs a provider block.
• How to make videos with Codex using GPT-6 Astra — the native-OpenAI version that needs no provider wiring.
• How to make videos with Claude Code — the Claude Code version of this setup.

Start making videos with Codex using Muse Spark Open

The setup is a one-time cost: install Codex, point it at Muse Spark Open, add the Pexo skill, and from then on every video is a conversation. Muse Spark Open handles the planning, and Pexo's AI video generator turns the plan into a finished cut with no timeline to operate and no model prompt to engineer. Open the social media video workspace and describe your first video.