Build it once, run it often

One image from one prompt does not need a canvas. The canvas starts paying when the same shape comes back: the same four angles of the next product, the same treatment for the next client, the same series next month.

On the canvas the prompt, the model, the references and your own style are blocks you connect. Save it, and running it again is changing one input.

A generator node shows the ports that model actually has, and greys out the ones it does not. Attach a file to a port the current model cannot read and the studio switches to a variant that can, or tells you that nothing does.

That is the same rule as on the make screen, from the same function. Two screens with two answers is how you end up paying for a run that quietly dropped half its input.

What the browser sends when you press Start is the flow and the node, nothing else. The server reads the saved graph, works out the inputs itself, and prices the run from the model in that graph.

That is a security decision rather than a design one, and it is why the amount on the button is the amount you pay.