--- title: "Production canvas" description: "The Marxx production canvas covers research, writing, image, video, audio, talent and publishing across roughly sixty node types, and creates Meta campaigns, ad sets and ads from the same graph, always paused for a human to review." canonical_url: "https://marxx.ai/features/production-canvas" --- # A production canvas that goes from concept to a paused Meta ad. Around sixty node types across research, writing, image, video, audio, talent and publishing, on one graph. The Meta campaign, ad set, ad and carousel card are nodes too, so an approved concept becomes a live-ready ad without an export step. **One graph, end to end.** Research, script, image, video, voice, talent and the Meta objects are all nodes, so nothing is exported, renamed or re-uploaded between steps. **Roughly sixty node types.** Web scraping and brand DNA through to lip sync, voice cloning, dubbing, motion control, AI avatars and UGC. **Published paused, always.** Campaign, ad set and ad are created in a paused state. Nothing starts spending because a graph finished running. ## Questions ### What is the production canvas? It is a node-based graph where each step of making an ad is a node you can wire to the next: research, script, image, video, voice, talent, then the Meta campaign, ad set, ad and carousel card. The whole path from concept to a ready-to-launch ad lives on one canvas instead of across a brief document, an editing tool and Ads Manager. ### What can the nodes actually do? Roughly sixty node types. Research and grounding: web scraping, HTML extraction, brand DNA and product DNA. Writing: text, script splitting and language transformation. Image: generation, editing, prompt-driven edits, background removal, face swap and layers. Video: generation, animation, motion control, extension, upscaling, resizing, trimming, combining, frame and audio extraction. Audio: voice generation, voice cloning, voice changing, dubbing, transcription and music. Talent: AI avatars, UGC studio and UGC rendering. Then publishing. ### How do I get variants without rebriefing? An approved concept fans out on the canvas by hook, format, avatar or language, so the variant set is generated from the same graph rather than written again from scratch. A router node lets one graph branch on a condition, which is how a single concept produces a different treatment per segment without a second canvas. ### Does it publish to Meta, or do I export files? It publishes. Campaign, ad set, ad, carousel card, placement selector and audience generator are nodes on the same graph, so an approved creative is built directly into the ad account. There is no export, no handoff, and no re-uploading a file someone renamed. ### Will it start spending money on its own? No. Everything published from the canvas is created paused. A human turns it on in the ad account, which means a graph that runs unattended, or an assistant that runs one for you, cannot start spend. Pushing an ad live stays a deliberate human action. ### Where do the concepts come from? From the intelligence, not from a blank page. A fatigue finding, a competitor teardown, a coverage gap or a persona objection opens on the canvas already loaded as a storyboard and script, so production starts from a reason someone can defend in a review. ### Can a human stay in the loop? Yes, and that is the intended shape. Nodes are editable at every stage, comment and note nodes let feedback live on the graph next to the asset it is about, nothing publishes without approval, and the graph is the record of what was decided. A review becomes a walk through the canvas rather than a thread of file versions. ### Is there quality control on generated assets? There is a review step rather than blind acceptance: generated assets are checked against the brief before they move down the graph, and a failed check is regenerated selectively rather than restarting the whole run. Long jobs like video and lip sync run as jobs you can watch, so a five-minute render does not block the canvas. ### Can I drive the canvas from Claude instead of the UI? Yes, for generation. The same generation nodes are exposed as tools over the Marxx MCP server, so an assistant can research, write, generate, chain one result into the next and check the output, with brand knowledge, personas and competitor intelligence available from the same connection. Publishing to a live ad account is deliberately not one of those tools. ### Is there a video editor? Yes. A timeline editor sits alongside the canvas, and canvas nodes contribute to it directly, so a clip generated on the graph can be trimmed and assembled without leaving the workspace or re-uploading anything. ### Does the loop close after publishing? It does. Once an ad is live it re-enters the same analysis as everything else in the account, so the next round of fatigue scoring and hook comparison includes the creative the canvas just produced. [Book a demo](https://marxx.ai/book-a-demo)