Lay out prompts, reference images and generation nodes on one canvas, connect them, and generate with GPT Image 2, Nanobanana and the site's other models, paid with your credits.
Prompts, references and results stay side by side.
A single prompt box only shows your latest attempt. On the canvas every prompt, reference image and result stays where you put it, so you can compare versions, reuse inputs and branch from any result.
Text nodes hold prompts and notes, image nodes hold uploads and results, and a generation node combines the text and images connected to it.
Type @ in a prompt to reference a connected image, text, video or audio node. Referenced images show as thumbnails right in the prompt.
Generating several images at once creates a stacked group. Expand it to compare and pick a main image; a line from the group passes every image on, and a line from one image passes only that one.
Crop, upscale, split, change the viewing angle or mask an area to redraw from an image's toolbar. Each result becomes a new node.
Create a text node for the prompt, then upload or drag in reference images. Each one becomes its own node.
Connect your text and images to a generation node, then pick a model, an aspect ratio and how many images to make.
Results appear as new image nodes. Download them, save them to My Assets, or connect a result to the next step to keep refining it.
Connect one result to several generation nodes to try different prompts or models on the same image.
A generation node lists its inputs; change their order to control which image the model treats as the main reference.
Paint a mask over part of an image and describe the change; the model receives the image and the mask as references.
Undo and redo, copy and paste nodes with their connections, box-select, and use the minimap on large canvases.
A line from a batch passes all of its images to the next node. Open the batch and connect from one image's own point to pass only that image; on a closed batch the line shows which images it carries.
Connect Codex or Claude Code to the Banana Art Lab MCP server and ask it to generate images onto a canvas by name. Its prompts and results arrive as nodes you keep editing here.
While you are signed in, canvases and My Assets sync to your account, so they open on any device where you sign in. You can also export a canvas as JSON and import it again.
Explore ideas with GPT Image 2 Saver at 2 credits an image, then run the prompt you like on GPT Image 2 at high quality or on Nano Banana Pro.
One product photo and a prompt for each scene, generated as a batch with GPT Image 2 Saver.
A character reference connected to several prompts, so the same character appears in every scene.
A first draft, then masked redraws and an upscale of the chosen version for print.
The plugin manager adds more node types, such as Markdown, SVG, HTML, 360° panoramas and sticky notes.
| Feature | Banana Art Lab canvas | Single prompt box |
|---|---|---|
| See every version side by side | - | |
| No coding required | ||
| Reuse prompts and references | - | |
| Reference several images with @ | - | |
| Batch generation | ||
| Edit results without leaving the page | - |
ComfyUI is a free, open-source node editor that runs open models such as Stable Diffusion and Flux on your own GPU. The Banana Art Lab canvas also connects nodes, but it runs in your browser with hosted models such as Nano Banana Pro and GPT Image 2: there is nothing to install, and you pay per image with credits.
Thousands of artists are already creating amazing workflows. Share yours and learn from the best.
Active Node Artists
Shared Workflows
Daily Creations
Open the canvas and start with your 15 free sign-up credits, enough for seven GPT Image 2 Saver images.