Ideogram Canvas

Connecting creative tools into workflows

Role
Design Engineer
Year
2026
Deliverables
Design, Front-end, Interactions
Type
Product
Stack
Figma, GPT-6 Astra, Claude Opus 5.5, Ideogram 4.6

I designed and engineered the Canvas frontend from the ground up. It brings images, creative apps, and an agent into one workspace, from a single generation to connected workflows and focused editing.

Hyper-realistic yearbook photo of a fictional, eccentric character in a comedic style; exaggerated facial expression, absurd hairstyle, strikingly inappropriate outfit, studio lighting, high school yearbook aesthetic, ultra-detailed, realistic skin texture, subtly humorous
Generate ImageGenerated yearbook portrait with oversized hair, glasses, floral jacket, and a blue tie3:4 · 1 imageGenerate Video3:4 · 1 video
Canvas
CanvasGrid
17,172 creditsShare
100%
Generate

The canvas

A place to start.

Canvas begins as a place to collect and arrange your work. Upload or drop in images, move them around, add text and notes, and group related ideas into sections.

I wanted the workspace to be useful from the first image. The familiar canvas interactions provide a foundation for the apps, workflows, and editing tools that come next.

Creative apps

Use a tool, get a result.

Image apps, video apps, and models are available from the floating toolbar. An app can be used on its own: add Generate Image, enter a prompt in the right panel, and create an image directly on the canvas.

The node keeps the output visible while the panel holds its settings. This gives each task a consistent place for its controls without filling the workspace with forms.

Generate Image
Output

Your result will appear here

1:1 · 1 image

Generate Image

Model
Auto
Reference images0/10
+
1:1
1
Advanced settings

Connected workflows

Connect the next step.

Each app can also become a step in a workflow. A prompt can feed several Generate Image nodes to explore different directions. An image can pass through background removal, upscaling, and a product mockup.

Dragging a connector into empty space opens a choice of compatible apps. Once connected, the path makes the relationship visible. Run a single app or start the workflow from its source, and keep the inputs and results together on the canvas.

Prompt nodes hold reusable instructions. Assistant nodes add an LLM step that reads instructions as part of the workflow.

The canvas agent

Work alongside an agent.

The agent offers another way to create. Open it from the left toolbar and describe what you want in natural language. Its results are placed on the canvas, where you can arrange them and continue working with apps.

The agent works alongside the workspace. You can move between a conversation, direct manipulation, and connected tools as the idea develops.

Focus Mode

Give one image your full attention.

Select an image and press Tab, or choose Focus from its toolbar. The image moves into the center of the browser as the surrounding panels enter. The workspace shifts from arranging many assets to refining one.

Apps and the agent are available here too, with edits directed at the focused image. New versions appear along the bottom, keeping earlier results within reach as you explore.

Annotations let you point to a specific part of the image and describe a change. Collect several comments, then send them to the agent together. When you are done, download the result or return to the canvas with the latest version.

Interaction exploration

Two images, a new direction.

In a one-off exploration, I tried making image generation a direct gesture: bring two images together to create something new.

The gooey interaction makes the merge feel tangible, visually connecting the two source images before revealing the result.

Design & engineering

Building the behavior was part of designing it.

I owned the product design, interaction design, and frontend engineering for Canvas, starting from an empty foundation. That meant designing both the overall workspace and the details that make it feel coherent: selection, connectors, floating panels, view changes, and generation states.

The product needed to support different levels of involvement. Someone could arrange images, use an app once, build a workflow, or ask the agent for help. I brought those ways of working into the same interface, then built the frontend interactions that connect them.

Focus Mode makes that approach especially visible: the transition establishes which image you are editing, the panels bring its tools into reach, and the version strip keeps the history close to the result.

The result

Room for an idea to develop.

Canvas brings creation, organization, and refinement into one workspace. An image can begin with a prompt, become part of a workflow, and move into focused editing without losing its place in the broader project.

20%
Drag to explore · ⌘ / Ctrl + scroll to zoom