add the option to import a high fidelity comp (like a screenshot of another website, or a mockup on dribbble) and build a website from that. I've been experimenting with this in the chatgpt ui, and it appears the latest models are capable of this. However it's a multi step operation and something needs to keep poking the model to get it to output the next section, and to perform other tasks (like using dalle to replicate image assets without violating anyone's copyright). Therefore a great use case for Assistants API and the super long context of gpt-4-1106-preview
which brings me to my next point: Assistants and function calling. Perhaps Assistants are too bloated and difficult to control... but the ability to call tools in a structured way (as can be done with any of the modern gpt 4 variants) will make multistep operations possible, without being attended by a human
brand definition step: a human web dev / ux designer building your site from scratch always ask for your logo before they build the page, so that they can pick suitable fonts and colors. If you do not have a logo, the human will usually insist on creating one and coming up with brand colors (if they don't, they should). Happily, if we map out a step by step process for the AI, there should be no problem using gpt to come up with this based on the name and one line description of the business
allow imports from common wire framing tools like balsamiq and draw.io ... and integrate with zapier to allow for universal integration between every tool somebody might use to make a wireframe, and your app
What are your priorities for feature development? I love this project and I would be happy to help you build it out into something that's commercially viable as a saas.
a few months ago I built an interactive webpage editor with gpt 4 as a proof of concept... start off with the source code of basic webpage like what your tool produces, and a system prompt that modifies the page and adds content by replying to human requests with JQUERY commands, which tweak the dom in real time. I never commercialized it because the 8k context window was a little tight... but with 128k context it's worth revisiting.
add the option to import a high fidelity comp (like a screenshot of another website, or a mockup on dribbble) and build a website from that. I've been experimenting with this in the chatgpt ui, and it appears the latest models are capable of this. However it's a multi step operation and something needs to keep poking the model to get it to output the next section, and to perform other tasks (like using dalle to replicate image assets without violating anyone's copyright). Therefore a great use case for Assistants API and the super long context of gpt-4-1106-preview
which brings me to my next point: Assistants and function calling. Perhaps Assistants are too bloated and difficult to control... but the ability to call tools in a structured way (as can be done with any of the modern gpt 4 variants) will make multistep operations possible, without being attended by a human
brand definition step: a human web dev / ux designer building your site from scratch always ask for your logo before they build the page, so that they can pick suitable fonts and colors. If you do not have a logo, the human will usually insist on creating one and coming up with brand colors (if they don't, they should). Happily, if we map out a step by step process for the AI, there should be no problem using gpt to come up with this based on the name and one line description of the business
allow imports from common wire framing tools like balsamiq and draw.io ... and integrate with zapier to allow for universal integration between every tool somebody might use to make a wireframe, and your app
What are your priorities for feature development? I love this project and I would be happy to help you build it out into something that's commercially viable as a saas.
Cheers :)