With the conference focused on “intelligent partners” and real-world deployment, FigCraft keeps complex visual production moving on a single canvas through task decomposition, model scheduling and node-based collaboration.
On 17 July, the 2026 World Artificial Intelligence Conference and High-Level Meeting on Global AI Governance opened in Shanghai under the theme “Intelligent Partners, Creating the Future Together”. Spread across the Expo, Zhangjiang and West Bund districts, the conference ran more than 140 forums, drew over 1,100 exhibiting companies, and presented more than 3,000 frontier results and over 300 global product debuts — framing AI’s shift from tool to collaborator within the shared context of industrial evolution and global governance.
At this juncture, the visual creation agent FigCraft released version 2.1.6. The update further establishes a video-task-oriented direction, combining task comprehension, step planning, model invocation and the intelligent canvas to help creators organise a continuous workflow from visual setup through to video output.
As image and video generation models keep advancing, AI creation is moving from “producing a result” to “completing a task”. Character design, scene building, storyboarding, keyframe generation, video production, asset revision and final assembly are all interdependent, and creators often switch repeatedly between models, platforms and folders. Seeing the structure of the task, managing intermediate results and maintaining continuity have become practical problems in AI video production.
The intelligent canvas: unfolding a complex task into an executable visual structure
FigCraft 2.1.6 carries this process on an intelligent canvas. Once the user states a complete creative goal, the agent interprets the task and plans the steps, turning characters, scenes, storyboards, keyframes and video segments into visible nodes on the canvas; the connections between nodes express asset inputs, task dependencies and the flow of results. An execution chain that used to be hidden across many separate operations becomes a visual structure that can be inspected and understood.
The canvas also serves as the human–agent interface. The agent decomposes the task and calls the appropriate image models, video models and creative tools, while the creator can follow the nodes to inspect the process, adjust local results, swap assets or change direction downstream. Multiple options for the same character or scene can be kept as parallel branches on the canvas, reducing the need to redo an entire pipeline because of one local change.
For video work this visibility matters especially. Images establish a relatively stable visual foundation for characters, scenes, style and keyframes, while video nodes build on them to form moving shots and continuous content. With upstream and downstream steps in one workspace, creators can judge more directly where an asset came from, where it will be used, and which later results a given change will affect.
The approach applies to AI comic drama, AI short drama, brand video and commercial visual content. In AI comic drama, for instance, a creator can continuously organise character settings, scene references, storyboard frames, keyframes and video segments on one canvas, so that scattered generative capabilities and assets gradually form a production process that can be inspected, adjusted and carried forward.
What FigCraft 2.1.6 explores is moving the intelligent canvas from a space that displays assets to one that executes tasks: goals are decomposed there, models and tools cooperate there, intermediate results are submitted to human judgement there, and the video takes shape there. At a moment when “intelligent partner” is the conference’s keyword, the update offers a more concrete product practice for how AI takes part in real creative work.