FreeClaude Code7 min read

Higgsfield's New CLI + Claude Code: Turn a Folder of Product Photos Into Ads on Autopilot

Higgsfield just shipped a CLI that plugs straight into Claude Code. Point it at a folder of product photos and it generates model try-on videos and cleans up dropshipping-grade images for you - one product or your whole catalog. Here is the exact workflow, plus the full video walkthrough.

Marcus Volsted
Marcus Volsted

AI & Web Consultant July 3, 2026

Higgsfield's New CLI + Claude Code: Turn a Folder of Product Photos Into Ads on Autopilot

Most people will scroll past this. That is the mistake.

Higgsfield just shipped a CLI, and the moment you connect it to Claude Code, your editor turns into an ad factory. You drop a folder of product photos in, describe what you want in plain English, and Claude does the tedious part - picking the model, wiring up the generation, and rendering video and images while you go do something else.

I recorded the whole thing end to end. In the walkthrough below I take a few flat product shots - a hoodie, a pair of tech pants, and a photo of a model - and turn them into a vertical video of the model actually wearing the clothes. No editor, no studio, no shoot.

The full Higgsfield x Claude Code walkthrough, start to finish.
馃幆 Result
Drop product photos in a folder, describe the shot, and let Claude drive Higgsfield. You get platform-ready video and images back without leaving your editor - and you can point the same setup at a hundred products instead of one.

Why this matters more than one nice video

One video is a party trick. The real unlock is scale.

Anyone can generate a single clip on a website. What you cannot do on a website is run the same job across your entire catalog without clicking yourself to death. The CLI is what makes it programmable. Once Claude Code can call Higgsfield, "make a try-on video of this product" becomes something you do once, or five hundred times, for the same effort.

Here is a real one. An e-commerce client had 40 low-quality, dropshipping-style supplier images - the kind of muddy, badly-lit photos that quietly kill trust on a product page. I used this exact CLI workflow to clean all 40 up. Actual hands-on time: under five minutes. The rest ran on its own.

40
supplier images fixed in one run
5 min
actual hands-on time
30+
AI models on tap
9:16
platform-ready by default

That is the difference between a toy and a tool. The toy makes one image. The tool makes your whole store look expensive while you get coffee.

Why run it inside VS Code and Claude Code

You can run Claude Code as a standalone app, but for this I use it inside Visual Studio Code. Two reasons.

  • You see what is happening. Every file Claude reads, every command it runs, every asset it drops in - it is all right there in front of you instead of hidden behind a chat window.
  • You can orchestrate files. This workflow is folders of images going in and finished media coming out. Having your files and your AI in the same window makes that natural.

If you have never set it up, it is three steps: download VS Code, open an empty folder for the project, and install the official Claude Code for VS Code extension by Anthropic. Log in through your browser once and you are ready.

Connecting Higgsfield to Claude Code

Higgsfield is a platform that sits on top of most of the frontier image and video models - so instead of juggling a separate account for each one, you get a single door to all of them. The CLI is how you hand that door to Claude.

You do not need to memorize any commands. Higgsfield gives you the exact install command on their platform. Copy it, paste it into Claude Code, and let Claude do the rest - it installs the CLI and prompts you to log into your Higgsfield account.

馃挕 Tip
Stuck on the setup screen? Take a screenshot of your Higgsfield dashboard, paste it straight into Claude Code, and ask it to install the tools for you. It reads the screen and figures out the commands. This is the quiet superpower of working in Claude Code - when you are lost, you show it what you see instead of typing out the problem.

Prefer the other route? Higgsfield also runs an MCP server if you would rather connect that way - I broke that path down in Generate Video and Images Without Leaving Your IDE. The CLI is the newer, more scriptable option, and the one I reach for when I want to batch.

Setting up the assets

The whole job starts with a folder. In the walkthrough I make a subfolder called hoodie-and-pants and drag in everything the model needs to work with:

  • The brown tech pants
  • The hoodie, shot front, back, and side
  • A photo of the model

That is it. The more angles you give it, the more the AI has to reason from - a front, back, and side of a garment is enough to keep the product consistent as the model turns.

The prompt that does the work

Now you just describe the outcome. I tell Claude what is in the folder and what I want out of it:

text
In this folder are photos of a hoodie (front, back, side), a pair of
brown tech pants, and a model. Generate a 10-15 second vertical (9:16)
video of the model wearing these exact clothes, on a simple light-gray
background, showing the outfit from a few different angles.

Notice what the prompt is doing. It names the assets, sets the format (vertical 9:16, so it is ready for Reels, TikTok, and Shorts without cropping), fixes the background so nothing distracts from the product, and asks for movement so it reads like a real shoot instead of a static photo.

Pick the right engine for the job

For the walkthrough I ask Claude to use a faster, cheaper model - Kling 3.0 fast - because I am demonstrating, not shipping. That is the move: draft on a cheap engine, finalize on a premium one.

馃挕 Tip
Treat engines like print quality. Rough out the idea on a fast, cheap model until the shot is right, then re-render the winner on a premium engine like Sora 2 or Veo 3.1. You pay for the good version once, instead of paying premium prices for every throwaway attempt.

Let it run on its own

Because this is a folder-in, media-out job with no decisions left to make, I set Claude to bypass permissions so it runs autonomously in the background until the task is done. No babysitting, no clicking "allow" every thirty seconds.

鈿狅笍 Heads up
Bypass permissions means Claude runs commands without asking each time. Only do this inside a dedicated project folder where you know exactly what it can touch - never in a folder full of important files or pointed at your whole system. Scope it, then let it run.

The result, and how to make it better

The video that comes back is not flawless - but for a simple prompt and the cheapest engine on the menu, it is genuinely good. The model wears the actual clothes, the angles change, and it looks like something you could put on a product page.

That is the floor, not the ceiling. Two easy levers push it higher:

  • Use a premium engine. Swap Kling 3.0 fast for Sora 2 or Veo 3.1 on the final render and the quality jumps.
  • Let an agent optimize the prompt. Instead of hand-tuning the wording, put a second agent in charge of rewriting and scoring the prompt until the output hits your bar. That is the looping idea applied to media - I broke the whole pattern down in Put Your AI in a Loop.

Where this actually pays off

The demo is a single outfit. The point is everything you can aim the same setup at once it is running:

  • Fix supplier images at scale - turn muddy dropshipping photos into clean, on-brand product shots, like the 40 I cleaned up in one run.
  • Spin up ad variations - generate ten versions of a creative in different angles, backgrounds, and formats, then test which one converts.
  • Replace the mini photoshoot - get try-on video and lifestyle shots without booking a model, a studio, or a photographer.
  • Stay consistent across a campaign - lock in one model or character and reuse the same face and look across every product.
  • Batch a whole catalog - one prompt, run across every SKU, while you work on something else.

The website generates one image. The CLI generates a workflow. Once your media lives in the same place as your code and your files, you stop making assets one at a time and start running them by the hundred.

That is the whole idea. Higgsfield brings the models, Claude Code brings the hands, and your editor quietly turns into the studio. Point it at one product to learn it - then point it at all of them.

The Higgsfield + Claude Code Prompt Pack

The exact copy-paste prompts from this workflow - product try-on video, batch image cleanup, and ad variations - plus the setup checklist. Enter your email and I'll send it over.

Want this running on your product catalog?

I set up AI media workflows that turn a folder of product photos into store-ready video and images - across your whole catalog, not one product at a time. If your product pages need to look expensive without an expensive shoot, that is what I do.

Tell me about your store