Image and video generation in Codex with Genvas
Genvas, a desktop app for AI images and video, ships a local MCP server that Codex can use to build boards, price them and run them. The results appear on the board in the app.
Genvas’s maker, Dee, wrote this page, including the last section on other tools.
On this page
Before you start
This page follows Genvas 1.3.9, read on 9 November 2026. You need the following:
- Genvas installed on macOS (Apple silicon), Windows 10 or 11 (x64) or Linux (.deb, x64). No Genvas account is required.
- Codex installed on the same computer.
- An API key from one of the providers Genvas lists. You pay that provider directly, and Genvas adds nothing on top.
Connect Codex and make an image and a clip
- Open Genvas. Codex can use it only while the app is running. If a licence screen appears, activate it first.
- Select Connect on the Codex row under Settings, General, Agents. The button then reads Connected. The same Connect button appears during onboarding.
- Check the config file. Connecting adds a
[mcp_servers.genvas]table to~/.codex/config.tomland installs a prompting skill namedgenvas-promptingfor Codex. Open the file and look for the table. - Add a key for one provider under Settings, Providers. The list is OpenRouter, OpenAI, Fish Audio, ElevenLabs, Google Gemini, Higgsfield, Kie.ai, fal.ai, xAI and Sume. A provider without a key shows No key. The panel says: “Keys stay on this computer. You pay each provider directly.”
- Set the Daily limit under Settings, Spending. The app says agent runs stop only at the daily limit. The setting “Ask before a run costs more than” applies only to runs you start in the app, so it does not stop Codex. Once spending over the last 24 hours reaches the limit, paid agent calls are refused.
- Start a new Codex session, so it begins after the config file changed.
- Ask Codex for an image in plain words. For example: “On a board called Lighthouse, make a wide hero image of a lighthouse at dusk. Price it first and wait for my OK before you run it.” Codex uses the board you name or creates one. It builds a prompt node, then a media node with a model node attached, and leaves a note with the goal and status.
- Request a video clip the same way. For example: “On the same board, make a short video clip of the lighthouse beam sweeping over the water. Price it first and wait for my OK.”
- Read the dry-run price. Codex prices each board with
run_workflowanddryRunset to true, which spends nothing. Compare the figure with your daily limit. If it is higher than you want, ask for a cheaper model, or for draft mode, which previews OpenAI images cheaply. - Approve the run by replying in plain words, such as “Go ahead”. Codex runs the board with
run_workflowand pollsrun_statusuntil the run is done. - Find the results on the board in Genvas. If you also want the files in a folder, set Settings, Export, Export folder. With no export folder set, nothing is exported.
If it does not work
Genvas is not open. Codex then gets the message “Genvas isn’t running” and is told to open Genvas.app. Open it, and if it shows a licence screen, activate it first.
Settings, Providers shows No key or Key rejected. No key means you have not added a key for that provider, so add one. Key rejected means the provider refused the key, so replace it with one the provider accepts.
Settings, General, Agents shows Not connected. Select Connect on the Codex row, wait for the button to read Connected, then start a new Codex session.
A run was interrupted. If Genvas is closed or restarted during a run, the run is marked as interrupted and the agent is told to run that node again. Reopen Genvas and let Codex rerun the node.
Other tools for images and video in Codex
Each description below comes from the tool’s own page, dated or read between 18 June 2026 and 10 November 2026, so check the page for current details.
OpenAI’s image generation in the ChatGPT desktop app uses gpt-image-2 and counts toward your general Codex usage limits. For larger batches you set OPENAI_API_KEY and ask for generation through the API, so API pricing applies.
Yir is a skill you add with npx skills add yir-ai/skills. One Yir key covers GPT Image, Nano Banana, Seedream, Seedance, Kling and others. It quotes a price first and stops at a cap you set. Codex blocks network access by default, so you approve it when asked or turn it on in config.
AnyCap installs for Codex with npx -y skills add anycap-ai/anycap -a codex -y. It gives Codex anycap image generate and anycap video generate commands under one login. Its guide is written in Chinese.
Medux is a remote MCP server that you add to Codex with a Medux API key. Its tutorial covers turning one source image into a video with the MiniMax H3 model.
Picsart for Codex is a plugin that gives Codex access to 140+ models for images, video and audio, paid in Picsart credits. It lets you review a credit estimate before anything is created.
Pexo makes whole videos from a plain-English description. It writes a shot script, picks a model for each shot, mixes a soundtrack and returns an MP4. You need a Pexo API key.
More guides & comparisons
ComfyUI vs Flora, Krea and Genvas for API models
ComfyUI, Flora, Krea Nodes and Genvas compared on price, setup and API models, with who each suits. Written by the maker of Genvas (free beta).
7 AI image apps that show the cost before a run, ranked
Seven apps ranked by how exactly they price a job before it runs: Genvas (free beta) prices a whole board; the rest quote credits or tokens.
Five Figma Weave alternatives without a monthly plan
Genvas, NodeTool, ComfyUI, Imaginode and Visual Sandbox compared with Figma Weave's monthly credits. Genvas is free during the beta; NodeTool and ComfyUI are free too.
Start your first board.
Add a key and make something in a minute.
Free during beta. Open source soon, under AGPL-3.0.