Support for uploading multiple images or provide detailed images to update the code

[Liao1]

Hi team, appreciate your work which is awesome!

Is your feature request related to a problem? Please describe.
I ran it locally and played with it using Claude Sonnet 3.5 API, I found that the quality of generated code depends on the complexity of the provided image, for example, if I provide the screenshot of the whole page, the generated code can only show main sections of the page, but if I provide a screenshot which only contains the first section of the page, there is a higher degree of restoration. I tried to update the site using different prompts, it works, but it also need a lot of efforts to find and describe the different of generated page and original page.

Describe the solution you'd like
So I'm thinking whether it helps or not to allow us to provide an image of a specific section of the site and update it accordingly. The whole process will be:

1. Provide the complete image and generate the basic structure of the code
2. Update the code using images of different sections to achieve higher degree of restoration.
3. Repeat step#2 until we get a satisfied version and update the page manually.

Describe alternatives you've considered
An alternative solution is to allow to provide multiple images to provide more information to the model, not sure which solution is better.

[Liao1]

Actually I tried to upload video but got errors, which is Exception in ASGI application, similar to some of the previous issues.

[abi]

Hi @Liao1 thanks for the suggestion. I have explored splitting a web page up in order to get better results. While it works okay, it's not as great as when a model fully generates the entire web page (it's more consistent with colors, fonts, etc.). The newer models like O1 and Gemini are getting much better at large web pages so in the near future, we'll switch to those models to solve this problem. Claude 4 is probably also coming soon and should be a lot better.