Where This Idea Came From

I was inspired by CEO Michale Tiffany's GitHub project called "A Particular Set of Skills". He wanted to create a way for  one AI assistant to ideate on a problem and a separate assistant (that's actually good at making images) to pick up on that conversation and then deliver a good image.

‍

‍Problem

‍I had a house renovation idea — widening the front dormer — that only existed in a conversation with ChatGPT. ChatGPT couldn't actually draw it because the free plan had run out of image credits. I also had scattered thoughts in Claude that were related to the renovation ideas. I need to merge it all into one folder and then use an image generator to make a good enough image I could bring to an architect, or just my dad to estimate costs.

‍

‍Steps to Solve

1. Gathered up all the scattered renovation notes — paint colors and room ideas from one conversation, dormer and porch ideas from another — into one file in Fulcra, so nothing lived in two different places anymore.
2. Set up a simple "in tray" and "out tray" in Fulcra: one spot for requests ("please make a picture of this"), and one spot for finished results. Any assistant can check either one.
3. Using Claude wrote up the dormer idea as a proper request and dropped it in the "in tray."
4. Connected an image-generating tool (FLUX) so an assistant could actually fulfill that request.
5. Hit a real snag on the first try — a technical setting was blocking the image tool from uploading or downloading anything. Worked around it for the moment by generating a version from a written description instead of editing the real house photo, and came back to fix the actual setting afterward.

‍

Solution

‍Once the in tray/out tray were set up and the image tool was connected, getting a finished picture came down to fixing that one technical snag:

• Confirmed, in a new conversation, that the setting had actually been fixed — it turned out the fix only applies to conversations started after you change it, not the one you're already in.
• Downloaded the actual finished picture (not just a link to it) and double-checked it was really the file it claimed to be, rather than just trusting the label.
• Filed the finished picture back into the Fulcra "out tray," attached to the original request, so anyone opening that request later sees the ask and the result together.
• Marked the image tool as a trusted source going forward, so future pictures don't need the same double-checking from scratch.
• Re-ran the test with a variation on the original need and say it repeat.

‍

Result

• The dormer idea that used to stall out at "ChatGPT ran out of credits" now has an actual picture attached to it, sitting right next to the rest of the house's renovation notes.
• The request and the picture both survived even after closing the conversation and starting a new one because everything lives in Fulcra, not in any one chat's memory.
• The whole setup is now reusable — any other renovation idea sitting in the notes (a paint color, one of the other dormer options) can go through the same in tray → picture → out tray process without rebuilding anything.

None of this required Claude to learn how to draw. Rather it just required a shared drop-off spot in Fulcra, a tool that could actually generate images, and a bit of persistence working through one stubborn technical setting.

The result: an idea that started as a paragraph of text from a completely different AI tools ended up as a real, finished picture — filed and ready the next time anyone opens the house notes.