AI infinite canvas for reference images and visual workflows

An AI infinite canvas keeps reference images, notes, generated results and a creative assistant in one zoomable workspace. Keep the original, compare versions side by side and continue from a selected image when your visual project needs several rounds of refinement. Turn an image into a draggable orbit view. Choose Orbit Light (MiniMax H3: 768P / 2K) or Orbit Plus (Seedance 2.0: 720P / 1080P) for a silent five-second video.

Open tool

Updated Oct 15, 2026

What you provide: Uploaded, pasted or dropped images, saved library images, creative notes and images from an existing conversation.
What you get: A private canvas of images and notes, new images connected to their sources, a PNG of the selection or whole board, and a ZIP of selected originals. You can also analyze one image and extract selected elements as separate transparent PNG assets.

References: Up to 6 images per generation
Export: Layout · PNG / Selected originals · ZIP
Retention: 30 days from first addition to that canvas
Cost: AI generation uses the selected model’s Credits

What you can make

Compare creative directions

Keep the original, alternatives and feedback together to decide which subject, layout or palette to develop.

Extend one image into a set

Select a source image and use “Generate a content set” for product detail images, multiple views or a social media series.

Organize images from a chat

Use the chat’s canvas import action to arrange available images and their generation relationships in a browsable space.

Extract assets from an image

Extract visible objects, text blocks and decorations into separate PNGs while preserving source pixels and connections.

How to use it

Right-click a conversation in the assistant history and choose Import to canvas; on mobile, use the More menu beside the conversation.
Open the canvas and sign in to create or open a project. Upload, paste or drop images, add library items, or import images from an existing chat.
Drag blank space to pan and use the wheel to zoom. Move images, add notes and use box selection or Shift-click to arrange several elements together.
Select up to 6 reference images and use reference generation. Describe what to preserve and what to change, review the model and Credits, then generate. The original stays on the board.
For a longer workflow, send selected images and notes to the canvas assistant. To extend one source into a set, choose a purpose, image count and additional requirements. Click an output to inspect its direct predecessors.
Connect two images, check images 1 and 2 in the dialog, describe how to combine them, use a style, compare or describe both, then select Run with assistant. You can also select two images and choose Ask assistant, or reopen an existing manual link.
Select one image and choose Extract elements, add optional instructions and select Analyze elements. Review the list and regions, select the elements you need, then extract them. Progress is saved per element; successful assets appear on the canvas and unfinished items can be retried.
Double-click an image, choose View original, or select it and press Space to inspect the original at 1:1, zoom and download. Check the actual images and their sources, then confirm the cloud save. Export selected elements, or deselect everything to export the whole board. Download needed images before they expire. To download multiple originals, box-select images or hold Shift while selecting, then choose Download selected to save a ZIP.
The canvas loads original images directly. Zoom in to inspect details. Create a read-only link to this canvas version and see its visits.
Generate 360° view: Turn an image into a draggable orbit view.

Before you start

Images expire 30 days after first being added to that canvas. Reopening, saving, duplicating layers or restoring a layout version does not renew them.
AI generation uses Credits. Check small details and consistency across outputs; each request accepts at most 6 references.
Extraction prefers transparent backgrounds and supports up to 50 elements per analysis. Sources must be within 16 MiB and about 25 million pixels, subject to available processing capacity. Complex edges or overlaps may fail. Hidden parts are not reconstructed; text remains a pixel image, not editable type, PSD or vector layers.
ZIP downloads include only selected original images, exclude notes, and support up to 256 MiB per batch. Unavailable images are reported separately.
AI invents unseen sides. A full turn and seamless loop are not guaranteed; this is a video view, not a 3D model. Private results are kept for 30 days. Download the video or individual views before expiry. 360° view is available in the private workspace; interactive media is not included in shared snapshots.

Start with an example

Turn a product photo into a new scene

Add and select a product photo you have permission to use, then enter this in reference generation.

Generation prompt: Keep the travel mug’s visible shape, lid, color and markings from the reference. Place it on a wooden desk in soft morning light, with clear space for a headline. Generate 1 image. Do not add accessories or product specifications that are absent from the source.

Check the result: Compare the proportions, markings and lid against the source. Continue from a chosen original or output when revising; a text response alone does not prove an image was generated.

Open tool

Common questions

How is an infinite canvas different from an image generator?

An image generator handles an individual request. A canvas keeps multiple assets, versions, notes and source relationships together for comparison, organization and continued editing.

Can I use the canvas as a moodboard?

Yes. Arrange references, add notes about color or composition and export your selection or the whole board. Generate additional images only when you need another visual direction.

Does hiding connections remove the reference relationships?

No. You can show all connections, only the selected image’s direct predecessors, or no connections. These display choices do not delete images or relationships.

How can I download several original images at once?

Box-select images or hold Shift while selecting, then choose Download selected. The ZIP preserves the original PNG, JPEG, WebP or GIF files and excludes notes. Unavailable images are listed separately. Each batch supports up to 256 MiB of originals; downloading does not extend the 30-day retention period.

How can I quickly view an original image?

Double-click an image or select it and press Space. You can also use its expand button, the selection toolbar or View original in the context menu. The viewer loads the original resolution, with fit, 1:1, zoom and download controls. Press Esc to return to the canvas.

How can I share this canvas version and see visits?

Anyone with the link can view and download this version’s visible originals and notes without signing in. Later edits do not change it. Links expire with the images and can be revoked anytime. Each displayed page visit counts once. Refreshing or reopening counts again. Image loads and request retries do not add visits; this is not a unique visitor count.

Does a manual link run a task automatically?

No. The dialog asks for your intent first. Confirming sends both original images and your instructions to the assistant; cancelling keeps only the link. Swap image numbers to clarify the subject, scene or style source. The describe and compare presets request text only. A running assistant task continues, and the new request is queued.

How do I separate image elements into individual assets?

Choose Extract elements to analyze the image, then select elements to extract. Analysis uses the current assistant model and existing billing rules. Extraction uses source pixels and local matting without an image generation model. Successful results are separate transparent PNGs that can be moved, downloaded or saved together as a ZIP. Check their edges and completeness; hidden content cannot be recovered.