Auto drop images!!!

[unphased]

Issue

I would like to streamline aider workflow slightly

currently the way is

• screenshot (mac cmd+shift+opt+4),
• run /paste,
• after the next message we type /drop clipboard
• tab complete it, hit enter, to drop the image

the big friction point is having to manually remove the image being active in the chat, because the common case is to discard it. I would like to have a hotkey i can hit to automate /drop for something matching a pattern!

Even better! By default drop any images. If we want it back we can easily copy the file path that was shown in the earlier prompt in the terminal... or since with official flow it's still potentially in the clipboard one could just re-run /paste!

So then for the streamlined workflow what I would love to have is:

• auto drop images after one prompt to aider with the image
• a hotkey we can type to run /paste to slurp up image immediately without typing /paste. no biggie since user can /drop it manually if it was accidental. Bonus points if we can hit the key and make it do this to add the image from clipboard while we are already in the middle of prompt editing.

[5ocworkshop]

I've had a similar thought, I like your idea. Maybe instead of auto-drop we could confirm to drop when the next prompt is submitted and an image that was already submitted is still included in the chat? Basically the same effect but if you did intend to leave it, you can.

[unphased]

I think that's good. It does make me wonder, if images included in past prompts carry forward in the chat history or not. I've simply never had a need to re-submit the same image but maybe depending on the model it could be something that is needed sometimes!

[5ocworkshop]

My understanding is that all files loaded in the current chat will be re-submitted with each query, including the images.

[unphased]

what i was asking about was whether images could be included as part of "chat history" even if it is removed from the list of context files. It is doubtful since each MLLM has its own specific tokenization schemes for images and i assume chat history is assembled into one big string/file.

[5ocworkshop]

That is an interesting question, I don't have any visibility in to that one but am now interested in learning the answer :)

I also need to turn on --verbose and see if I can get a sense of how the files you include are ordered when submitted.