Make AI Videos That Actually Look Like You (Higgsfield + Claude Code Tutorial)

This shot would normally need a drone pilot, a VFX team, and a big budget. I was able to do it in half an hour with Higgsfield credits and Claude Code, and the character actually looks like me. So, let me walk you through how you can set it up yourself. To make this, you'll need three things: a Higgsfield account, and this video used about 280 credits, a Claude Code account, and six to eight selfies taken from different angles in natural light with no filters. The first thing that we need to do is connect the Higgsfield to Claude Code.

And to do that, you just go to settings, connect this, discover, and then connect. The reason why I'm not using the Higgsfield website is that the prompts to create this kind of video are really long. They include very precise timings for each second of the shot, camera movements, color rules, and a whole list of other things. And doing all of that on the website means that you have to paste huge blocks of text, choose the model again and again, change settings, and uploading the same four photos again for every single generation. So, that's why I prefer to connect Higgsfield to Claude account.

When you connect Higgsfield to Claude Code, you're still going to use your Higgsfield account and all of the same models, but Claude will do all of the clicking for you. You just need to describe what you want in normal language, and Claude is able to write the detailed prompt, upload your photos, choose the settings, run the generation, and save the finished files into a folder on your computer. And if something doesn't look right, you can just tell it what to change instead of starting everything from scratch. The second step is the most important one in the whole process, because this is where we make sure that AI actually looks like you in every shot. What we are making is something called a character sheet.

It's basically a casting photo for your AI double, showing your face and your outfit from three different angles. You need to put your photos into a folder that Cloud has access to and just ask it to make a character sheet with GPT image 2.5 showing you from the front, from the back, and a three-quarter portrait on a white background with soft natural light. And say, "Here's my outfit and please do not beautify me." And Cloud then turns that into a full detail prompt, [music] uploads all of your photos, and runs it in Higgsfield.

>> [music] >> And when the results come back, you can choose the one that looks most like you. And it's really important to get this part right because every clip that you make after is going to be built on top of it. And I'll be honest, my first attempts looked more like a glamorous older cousin than like me, but what fixed it was adding more photos from different angles and switching to more natural light. And adding that line to not beautify me and not change my age. And all of that made such a big difference.

Now that we've got a character that looks like us, we can start making the video itself. It's made out of two separate clips. For the first one, I describe the scene to Cloud the way I would explain it to a film director. And here's my prompt. Make an 8-second video, use Seedance 2.

5 video model. And I've tried a few other ones, but Seedance is the best one. In that video, I'm walking down this gray New York street looking bored. And then I slide my sunglasses down and say, "Let's fix this." And the camera flies through the lens into a pink city.

Use my selfies, the character sheet, and my street photos as a reference. If you look at the prompt that Claude writes from that, you will see it breaks the 8 seconds down into time moments, which is what makes the clip feel so controlled. It also adds a bunch of rules like keeping the whole city completely gray so that only my outfit pops and keeping it as one continuous shot with no cuts. The second clip is where the fun happens, and this is roughly what I asked for. Now, make a 15-second clip in one take filmed like an FPV drone following me.

A pink wave spreads out from me and transforms everything it touches. Taxis turn pink, steam turns into cotton candy become and pigeons become flamingos. Only show my face one, and when the camera spins round me in the crosswalk, and I wink. Claude also adds a long list of things to avoid such as cats, a flat pink filter. Claude will add all of the necessary things.

You can run two videos at the same time when you're doing it through Claude Code, but I would say make the first one and see how it performed. Check if you're happy with how you look and all of the camera movements, and then I would do the second clip. So, you can avoid losing credits. And the last step is joining the two clips together, and you don't actually need CapCut for it. You can just ask Claude to do it for you.

The reason why it looks so smooth is that the cut is right as the camera goes through the lens. And once you made one, it's really easy to make it your own because you only need to change the color and what it turns into. So, you could do gold lenses that turn the whole city to gold or blue lenses that flood the streets with water, green lenses where the city becomes a jungle. I mean, there are so many possibilities. On the 1st of November, I'm kicking off a 2-week sprint where you'll get videos and tasks to do each day.

And this will be focused not only on making videos like this, but also editing long-form videos, short-form videos, and a bunch of other things. So, if you would like to join that, I've put the link in the comments.