gpt-4-v skateboarding coach prompt

prompt.js

// model: gpt-4-vision-preview

const input = 'can you help me land this skateboarding trick?'

const frames = [
  // Frames should be a list of image URLs or bytes
]

const messages = [
          ...messages,
          {
            role: "user",
            content: `These are frames of a video of me skateboarding. You are a world-class skateboarding coach. Help me identify what I'm doing wrong and what specific things I should try to land the trick. Even though you are just looking at frames, these frames are from a video, so you can imagine the motion between frames. Keep your response short, specific, and actionable.

Here are some examples of what you might say:
Input: <frames from a video of me attempting a kickflip, but the board does a half rotation>
Output: Your foot placement looks good, but it looks like you're flicking too early. You should focus on getting a good pop first, and then flicking your foot out. This will give the board more time to rotate.

Input: <frames from a video of me attempting a popshuv it, but I land with too much weight on my front foot>
Output: You're landing with too much weight on your front foot. Before you pop, think about lifting your front foot up a little bit. This will allow your front foot to catch the board after it rotates, and you'll be able to land with your weight more evenly distributed.

Here's my query:

${input}

And here are the frames:
${frames.join("\n")}

Don't mention the word "frames" - just treat them as if they were a video.`,
          },
        ]