Hi, I am Chandima, working as a full-stack developer with Dotnet and Angular. I found your effort on this project very thrilling, and I enjoyed a lot playing around with it.
I have made some improvements that I felt were nice to have. But as I am not an expert on NextJs, I might have made mistakes in the framework usage and best practices. Please add any comments you might see that I can work with and improve so I can correct them.
These are the features I have added to this PR.
Even though the drawing is the same, clicking the "make real" button multiple times made different HTML designs. This prevented the user from improving the same design more and coming up with a better-fitting design. I have sent the API the current HTML generated and updated the prompt to have that HTML and upgrade it based on the user's drawing.
Note : This might be a bit costly as the token count is going to be high with bigger designs. So maybe later we can let the user decide whether they need to continue with the same design. Or if GPT implements a solution to maintain consistency, it would be a very nice one to have. But for the time being, addition 3 will be helpful to prevent this
No matter how clear the design drawings are, the GPT will mess up the design and create something not requested. (Vertical and horizontal placements are messed up, etc...) So I added an additional guides section so the user can give specific guides if the GPT keeps failing at a certain design.
If the user feels like the design is completely off and not getting any better, there is an option to select a checkbox and let GPT know to trash the current HTML and go with a new one.
Option 2 and 3 are getting reset after a successful generation since they have already catered the instructions into the HTML and will no longer be needed.
Hi, I am Chandima, working as a full-stack developer with Dotnet and Angular. I found your effort on this project very thrilling, and I enjoyed a lot playing around with it.
I have made some improvements that I felt were nice to have. But as I am not an expert on NextJs, I might have made mistakes in the framework usage and best practices. Please add any comments you might see that I can work with and improve so I can correct them.
These are the features I have added to this PR.
Note : This might be a bit costly as the token count is going to be high with bigger designs. So maybe later we can let the user decide whether they need to continue with the same design. Or if GPT implements a solution to maintain consistency, it would be a very nice one to have. But for the time being, addition 3 will be helpful to prevent this
No matter how clear the design drawings are, the GPT will mess up the design and create something not requested. (Vertical and horizontal placements are messed up, etc...) So I added an additional guides section so the user can give specific guides if the GPT keeps failing at a certain design.
If the user feels like the design is completely off and not getting any better, there is an option to select a checkbox and let GPT know to trash the current HTML and go with a new one.
Option 2 and 3 are getting reset after a successful generation since they have already catered the instructions into the HTML and will no longer be needed.
This is the feature in action.
https://github.com/SawyerHood/draw-a-ui/assets/15955557/c7f18ddd-b200-407b-966e-128efbb33a53
I hope you find these improvements helpful to the users and a good addition to your repo.
Have a nice day!