AI Copilot: How the Right-Side Chat Panel Drives Canvas Work
Use the real ChatPanel preview to understand how Copilot and Video Agent fit into canvas creation.
AI Copilot is not a separate chat page. It lives in the right-side panel of the main canvas, where prompt input, asset mentions, session controls, and generation confirmation all stay close to the workflow.
What this panel actually does
The current ChatPanel frontend has two working modes:
- Video Agent for plan building and staged video confirmation
- Copilot for natural-language edits, quick node creation, and creative orchestration
So this article is not really about “chat history.” It is about how you keep issuing instructions beside the canvas and let the panel turn them into actionable canvas work.
What Copilot is good at
Copilot works best when you already have a direction, some references, or a partial canvas and want to move faster:
- create a first draft on the canvas without manually building every node
- extend an existing branch with one more instruction
- reorganize reference assets into a new generation direction
- add the next image or video step without leaving the current canvas context
In practice, it behaves more like a working orchestration assistant than a generic chatbot.
The real interactions inside the panel
The actual ChatPanel UI breaks down into three layers:
1. Top area: mode and session controls
The header is where you switch modes, start a new session, dock the panel, or collapse it. In other words, Copilot is embedded in the canvas workflow rather than hidden behind another page.
2. Middle area: starter state, replies, and confirmation cards
In Copilot mode, an empty session starts with prompt starters. Once you begin using it, replies appear in the same stream; if a step needs explicit approval before generation, the confirmation UI shows up inline rather than in a disconnected flow.
3. Bottom area: the prompt editor
The input area is not a plain textarea. It is a rich prompt editor designed for references and mentions. The most important behaviors are:
- type natural-language instructions for what to create or modify
- use
@to mention assets, and in the full canvas workflow you can also mention canvas nodes
This article preview mounts the real frontend ChatPanel, so you can inspect the actual layout, tabs, fee indicator, and editor styling. To keep the article demo safe, submit actions are disabled while input and mention interactions remain available.
Why points and confirmation still appear in Copilot
Copilot does more than rewrite a prompt. When it is ready to turn a plan into real generation work, the frontend enters a confirmation step:
- it shows what still needs action
- it exposes point-related information for generation work
- it lets you decide whether to continue
That separation matters because creative orchestration and billable generation are not the same step.
Suggested usage pattern
If you want more reliable results from Copilot, a good rhythm is:
- keep each turn focused on one concrete goal
- mention assets with
@before asking for changes - let Copilot propose structure or direction first, then generate
- use the confirmation UI for cost-sensitive steps instead of chaining blind requests
Used that way, the ChatPanel feels less like a black-box chat window and more like a controllable production surface attached to the canvas.