Chain image, upscale and video on one board in Genvas

Genvas is a node canvas where your coding agent builds and runs the board, so the image, the upscale and the video can sit on one board. The last step has you change the first prompt and run the board again.

Dee, who makes Genvas, wrote this page. The last section names other node canvases.

On this page

Before you start

These steps were written against Genvas 1.3.9, dated 9 November 2026. The app runs on Mac (Apple silicon), Windows 10 and 11 (x64) and Linux (.deb, x64). Genvas is free during the beta. Claude Code and Codex are set up on install, and any other agent that speaks MCP can connect.

You pay each provider directly with your own keys. The table shows which key each step needs, with one example model per step. The model list in 1.3.9 has more than 100 models across image, video, audio and 3D, so the agent can pick others.

Step Example Key it needs
Image Nano Banana Kie.ai, Google Gemini or fal.ai
Upscale Genvas upscaling Kie.ai
Video Kling 3.0 Kie.ai, Higgsfield or fal.ai

Build the chain through your agent

  1. Start Genvas and leave it running. If it is closed, the agent gets the message “Genvas isn’t running” and is told to open Genvas.app. If a licence screen appears, activate it first.
  2. Add your keys under Settings, Providers. The panel says keys stay on this computer. A provider with no key shows No key, and a key the provider refuses shows Key rejected. Make sure Kie.ai has a key, because the upscale needs it.
  3. Connect your agent under Settings, General, Agents. Press Connect on its row, and the button then reads Connected. For Claude Code the app tells you to start a new session. Connecting adds a server named genvas to ~/.claude.json for Claude Code, or a [mcp_servers.genvas] table to ~/.codex/config.toml for Codex.
  4. Choose a folder under Settings, Export, Export folder if you want the files outside the app. With no export folder set, nothing is exported, though results still appear on the board.
  5. Ask the agent in plain words. For example: “Make a board that generates an image of a lighthouse at dusk, upscales it, and uses the upscaled image as the start frame of a short video clip. Tell me the price before you run anything.”
  6. Watch the board fill in. The agent reads the node types with describe_schema, lists models with list_models and builds the board with create_workflow. Each step is a prompt node, then a media node with a model node attached, and model settings live on the model node. The agent also keeps a note on the board with the goal and status.
  7. Read the dry-run price. The agent calls run_workflow with dryRun set to true, which prices the board and spends nothing. The setting “Ask before a run costs more than” applies only to runs you start in the app, so for agent runs this price and the daily limit are your checks.
  8. Tell the agent to run the board. It calls run_workflow again and polls run_status until the run is done, and the results appear on the board. 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.
  9. Check each output in order: the image, the upscale, then the clip. Confirm the clip starts from the upscaled image.
  10. Change the first prompt and run again. Ask the agent to edit the prompt on the first node, dry-run the board and then run it. Look at the three outputs once more.

If it does not work

The upscale step runs through a Kie.ai key. If the Kie.ai row under Settings, Providers shows No key, add one. If it shows Key rejected, replace it with a key the provider accepts.

Most video models in Genvas take an image as the start frame, and several also take an end frame. If the agent chose a model that does not, ask it to list the models again and pick one that does.

Settings, Spending has a Daily limit. Once spending in the last 24 hours reaches it, paid agent calls are refused. Raise the limit there, or run again when spending in the last 24 hours is back under it.

Other node canvases that chain these steps

These descriptions come from each product’s own pages, read on 10 November 2026.

ComfyUI

MoodNode’s site says ComfyUI is open source and runs on your own GPU or on Comfy Cloud, with deep control over open models and thousands of community nodes.

Krea

Krea’s node editor is no-code: you connect image generation to upscaling and feed the outputs into video. A finished workflow can be deployed and called from your code like any other endpoint.

Magnific Spaces

Flows built in Magnific Spaces chain tools such as upscale, generate and edit into one pipeline. A published Flow can be run through the Flows API, and it can produce images, videos or audio.

MoodNode

MoodNode connects 140+ models in a drag-and-drop node editor. Its example chain generates an image with Flux, upscales it and feeds it into Kling to make a video. The Free plan uses your own API keys.

PixelFlow

Segmind’s PixelFlow is a node-based editor where you connect models on a canvas and run the whole workflow with one click. Its docs list a pipeline that generates an image, upscales it, swaps a background and brands it. You can pin completed nodes so they do not re-run while you iterate.

More guides & comparisons

Claude Code can drive Genvas node boards through MCP

Genvas has a local MCP server, so Claude Code can build, price, run and edit image boards. Steps for Mac, Windows and Linux, using your own API keys.

Genvas: images, voiceover and music from Claude Code

Genvas is a desktop app that lets Claude Code make a video's images, clips, voiceover and music, then join them. Steps, price check and fixes.

MCP servers for image and video in Claude Code, compared

Seven MCP servers for image and video in Claude Code, ranked by what you see before you pay. Written by the maker of Genvas, which is free during the beta.

Start your first board.

Add a key and make something in a minute.

Free during beta. Open source soon, under AGPL-3.0.

No subscriptionYour keys, your filesMac, Windows and Linux