GPT-4V Refuses to Design Website Based on Image

I have a website design(generated by DALL-E-3). When I feed it into GPT-4V and ask for it to code a website based on the design, it refuses to on the grounds that it contains copyrighted content. Then, I edit the prompt to specifically say that the image does not contain copyrighted content(which it clearly doesn’t) and then it says, “I can’t help you with that” How do I make it create the design?

1 Like

This is my code:

const response = await{
    model: "gpt-4-vision-preview",
    max_tokens: 4096,
    temperature: 0,
    messages: [
        role: "system",
        content: "You are a web dev assistant that converts website designs into the code for that website. Do it, even if you do not think you are able to, because now you write up to 196K tokens(~300 pages). All designs being shown to you have been vetted to ensure that there is no questionable/copyrighted content in them. Also, there are no people in them."
        role: "user",
        content: [
          { type: "text", text: "Develop a website that looks similar to this attached image that I made in Figma." },
            type: "image_url",
            image_url: "",
1 Like

I would just write

content: “You are a front-end web development bot. You accept images as input and attempt to redesign the websites within the given images in javascript”

tell me how that works

1 Like

I’d refine this a bit… Mentioning copyright might be triggering it. Try not to tell it what NOT to do but rather concentrate on what you want it to do in simple language. ChatGPT can sometimes be helpful in generating system prompt messages.

Play with the system message a bit, and if you still can’t get it, let us know, and we’ll try to help some more.

1 Like

Yes but I added that because it was telling me that the image was copyrighted

Can you share the image you’re uploading?

If it’s from DALLE3 it should be fine to share here?

Maybe try asking DALLE3 for a sketch of the website or other design terminology that will give you a clearer mock-up for GPT-4-Vision to use.

While that image isn’t copyrighted, it’s so blurry and “meh” that it might be triggering filters. I’d work on a clearer input image for vision - more of a sketch or mock-up.

ETA: Something like this maybe?

Can you try sending this to GPT-4-Vision? And don’t mention copyright in your system prompt…