Loading an image at runtime still uses VRAM. In terms of VRAM usage, it's no different than having it stored as an animation frame. Construct loads it once to VRAM the same way all other textures are loaded to VRAM.
The only benefits of loading images at runtime instead of having them as part of the cap is it makes it quicker to preview and save, and if you want to update your game and save bandwidth/download time, the players don't have to re-download all the images they already have.
I know that, I was referring to a sort of "on-demand putting into VRAM." From the other topic which is kind of related to this issue, I've found out that everything is stored in RAM until it's loaded into VRAM when you load the layout. I'd like to be able to decide what gets loaded into/unloaded from VRAM better. My post was a little unclear, sorry.
manontherun - That's sounds like Asset Streaming. 3D games really use it. It loads on-demand what it needs. That's why there can be a little delay before your character model changes.
Unfortunately, Construct is very limited with how it can use resources and a lot of features that 3D game engines use (and can really be in 2D games too!), aren't in Construct. Construct was originally made to just run everything from a single .exe. It is not a good practice and a lot of people don't like it, so the whole encryption thing is a hot topic.