---
title: 'Treg: NEW Open Source Higgsfield is INSANE! 🤯'
source: 'https://youtube.com/watch?v=ukLfF99fHHI'
video_id: 'ukLfF99fHHI'
date: 2026-11-16
duration_sec: 1047
channel: 'Julian Goldie SEO'
---

# Treg: NEW Open Source Higgsfield is INSANE! 🤯

> Source: [Treg: NEW Open Source Higgsfield is INSANE! 🤯](https://youtube.com/watch?v=ukLfF99fHHI)

## Summary

This video introduces T-Reg, an open-source tool that lives inside Claude Code and Codex, which automates the creation of AI-generated UGC-style video ads. It walks through the four-step workflow—research, AI creator creation, voice selection, and video generation—and compares T-Reg to Higgsfield, a popular AI video app. The video also covers setup instructions, practical use cases for stores and agencies, and important ethical rules for using AI creators in ads.

### Key Points

- **T-Reg Overview** [00:01] — T-Reg is a tool catalog for AI agents, described as 'OpenRouter but for tools.' It provides one connection to over 4,100 tools from 127 providers, including video, image, voice models, and social trend data. It lives inside Claude Code or Codex, allowing users to generate videos through a single conversation.
- **Launch and Viral Post** [01:12] — On November 13th, Jason Joe, founder of Superdesign and AI Builder Club, posted on X calling T-Reg 'open-source Higgsfield.' The post highlighted its ability to replicate viral video formats, run inside Claude Code or Codex, and be open source. The demo video showed a giant cat walking down a city street with creators filming a selfie video, and it quickly gained tens of thousands of views.
- **AI UGC Video Skill** [02:55] — T-Reg ships a skill called 'AI UGC video,' which is a step-by-step playbook for creating user-generated content (UGC) style ads. UGC videos look like a normal person filmed them on their phone, which performs well in ads because they don't look like traditional advertisements.
- **Step 1: Research** [03:23] — The agent searches social platforms for the most liked posts in the user's niche and picks one as a reference. It then breaks down the video: what's on screen in the first 3 seconds, each beat second by second, camera movement, lighting, who talks, and when. This breakdown becomes the recipe for the new video, starting from proven content.
- **Step 2: AI Creator** [04:04] — The agent creates brand new characters for the roles in the reference video, building a character sheet with front, side, back, and close-up images. It also creates the location and props (like the product). This 'locking the assets' step ensures consistency across all takes, so the same person appears in every ad.
- **Step 3: Voice** [05:03] — The agent browses a library of over 1,000 English voices from Fish Audio. The user hears the line and picks the voice. The voice is handed to the video model as a reference for lip sync, as it sells the character as much as the face does.
- **Step 4: Video Generation** [05:30] — The agent sends everything—reference video, character sheets, location, product, and voices—to Seedance 2.5, Bite Dance's video model. The output is a video in the same format as the viral original but with the user's own creators, product, and words. The demo showed three versions of the same video in different settings (purple, city street, golden hour rooftop) for A/B testing.
- **Built-in Rules** [06:14] — The skill includes rules: the agent writes camera details first in every prompt, asks for real flaws (handheld wobble, focus hunting, skin texture), makes a low-resolution draft first, and honestly states it can't hear audio, so users must check voices themselves.
- **T-Reg vs. Higgsfield** [07:42] — Higgsfield is a website where users click around; T-Reg lives inside an AI agent. Higgsfield makes the video but leaves research, script, and characters to the user; T-Reg runs the whole job. T-Reg reaches over 4,000 tools, while Higgsfield is a video/image studio. Both use similar video models (Seedance 2.5), but T-Reg is open source and uses the user's own API keys first. Higgsfield wins on simplicity for those who prefer a polished interface.
- **Old Way vs. New Way** [10:10] — The old way: scroll for hours, message creators, wait for replies, send product, wait for filming, request changes, and get one version. The new way: tell the agent your niche and product, it finds a winning format, makes the AI creator, picks the voice, writes the plan, makes a draft, and produces three versions to test in one afternoon.
- **Use Cases for Stores and Agencies** [11:06] — For stores: start with a best-selling product, ask for a natural phone-style video with a handheld wobble and window light. For agencies: offer a new service by breaking down a client's winning ad, creating a brand creator, and delivering three versions. Also, use the research step for client calls to know what's working in their market.
- **Setup Instructions** [13:11] — Step 1: Open Claude Code or Codex. Step 2: Paste a message to read T-Reg's setup instructions and sign in (GitHub, Google, or email). Step 3: Paste a second message to follow the AI UGC skill. Step 4: Answer the agent's questions about niche, product, characters, and voices. The agent shows a plan for approval before making any video.
- **Doubts and Rules** [14:26] — Addressing doubts: no coding required (just paste two messages and answer questions); it's ethical to use AI creators if you remix the structure, not the people (no real faces, voices, or celebrity sound-alikes); label AI videos as AI and never present AI creators as real customers. The format comes from what's already winning in your niche.
- **Action Plan** [15:40] — This week: pick your best-selling product or best client, set up T-Reg with the two messages, ask the agent to find and break down the top video in your niche, and make one AI creator, one location, and one video to test against your current best ad.

## Transcript

Higsfield and it lives inside Claude Code. It's called [music] Tre. You type one message to your AI agent and it goes and finds a video format that's already winning in your niche. Then it creates an AI creator for your brand with a
face, an outfit, and a place to film. It picks a voice for them. And then it makes the finished video with the same top video models the big apps use. all from one conversation inside the AI you already work with every day. For an
online store, that's product videos that look like a real person filmed them on their phone. For an agency, that's a whole batch of ad videos for a client in three different versions you can test against each other. In this video, I'll
show you what TEG is, the four steps it runs, exactly how it stacks up against Higsfield, and how to set it up yourself. And stay with me because there's one rule built into Tre's own video playbook that keeps your ads on
the right side of the line and most people making AI ads right now are ignoring it. Let's get into it. Let me start with what actually happened. On the 13th of November, Jason Joe posted a launch on X. Jason is the founder behind
Superdesign and the AI builder club and his team builds tools for AI agents. The post called it open-source Higsfield. It said three things. It replicates any
viral video format. It runs directly inside Claude code or codeex and it's open source. The video in that post shows the whole thing working. A giant cat walking down a city street with two creators filming a selfie video in front
of it. Then the steps behind it. Finding the trending content. Filtering creators. Building an AI creator from scratch. Picking a voice from a library. and then a side byside of an original video and the new version shot forshot.
Within hours, tens of thousands of people had watched it. So, what is T-REG? In normal words, T-reg is a tool catalog for your AI agent. The team describes it as Open Router, but for tools instead of models. If you haven't
heard of Open Router, it's one connection that lets you reach lots of different AI models. T-REG does the same thing for tools. One connection and your agent can reach over 4,100 tools from 127 providers. Video models,
image models, voice models, social trend data, research tools. You don't sign up to each provider. You don't juggle 20 login. Your agent just asks for the job it needs done and T-REG finds the tool that does it. And because the code is
public on GitHub, anyone can read it, check it, or even run their own copy. The part this launch is about is AI video. Craig ships a skill called AI UGC
video. Craig ships a skill called AI UGC video. UGC means user generated content. a normal person filmed it on their phone. A creator holding your product, talking straight to the camera, walking down the street. That's the style that
does so well in ads right now because it doesn't look like an ad. And the skill is a playbook your agent follows step by step so it makes those videos properly instead of guessing. Here's how the flow works. There are four steps and your
agent runs every one of them from inside Claude code. Step one is research. You tell your agent your niche, say street style fashion. The agent searches social platforms for the most liked posts in that niche and picks one as the
reference. Then it breaks that video down. what's on screen in the first 3 down. what's on screen in the first 3 seconds, each beat second by second, how the camera moves, where the light comes from, who talks, and when. That
breakdown becomes the recipe for your video. And this is the smart bit. A post that already went viral tells you the framing, the pace, and the kind of people that work in your niche. So, you're not guessing what might work.
You're starting from something people have already proven they love. Step two is your AI creator. The agent makes brand new characters for the roles in the reference video. Not the real people, new people made from scratch.
For each one, it builds a character sheet. That's one image with the person from the front, the side, the back, and close-ups of the face. Then, it makes the empty location like a street or a kitchen. And then any props, like your
product. Tre calls this locking the assets. Here's why that matters. Video takes much longer to make than a still image. So you settle every face, the set, and the product as simple images first. Then you hand all of those images
to the video model as references. Every take after that keeps the same people in the same place. That's how you get a creator who looks like the same person in every single ad instead of a slightly different face every time. Step three is
the voice. The agent browses a library of over a thousand English voices from Fish Audio. You hear the line, you pick the voice. For the giant cat video, they picked a voice called Chloe for one creator and another voice for her
friend. The voice gets its own step because it sells the character as much as the face does. The agent hands that voice to the video model as a reference, so the lip sync follows it. Step four is
the video. The agent sends everything to Seance 2.5. That's Bite Dance's video model, and it's one of the strongest video models out there right now. The reference video, the character sheets, the location, the product, and the
voices all go in together, and out comes your video. Same format as the viral original. Your own creators, your own product, your own words. In the demo, they did something really useful. Next, they made three versions of the same
video. One in a purple setting, one on a city street, one at golden hour on a rooftop. Same format, three looks. So, you can run all three as ads and see which one people stop for. The skill also has rules built in, and they're
good ones. The agent writes the camera details first in every prompt because that's what makes AI video look real. It asks for real flaws on purpose, like a little handheld wobble, focus that hunts for a second, and visible skin texture.
It makes a quick lowresolution draft first and only makes the full quality version once you're happy with the draft. And it tells you honestly that it can't hear audio, so you check the voices yourself before the final
version. If you want to set up TREG and this exact video workflow for your own store or your clients, that's what we do every day inside the AI profit boardroom. You get the AI video ad system with the exact messages to send
your agent for research, AI creators, voices, and the final video. You get the step-by-step setup for Tre inside Claude Code, so your agent is making videos on day one. You get the full agent OS, so your agents already know your products,
your brand, and your clients before they make anything. You get a 30-day roadmap for using AI in your business, step by step, even if you've never automated anything before. You get four live coaching calls every week where you can
bring your exact product and your exact ad and get help with it. And you join over 3,400 business owners who are using AI to grow their businesses right now. description or go to aiprofitboardroom.com.
Now, the big question, how does this compare to Higsfield? Because Jason literally called it the open-source Higsfield. Higsfield is the popular AI video app. You log into their website, you pick a preset, you upload an image,
and you click generate. It has camera moves you can choose from, character tools, and it gives you access to lots of the top video models in one place. and it's really good at what it does. So, let me be fair to both. First, where
So, let me be fair to both. First, where you work. Higsfield is a website. You go to it and you click around inside it. T-reg lives inside your AI agent. You never leave Claude Code or Codeex. You just talk to your agent like you'd talk
to a video editor on your team. Second, how much of the job it does. Hicksfield makes the video when you ask for it. The research, the script, the characters, and the plan are mostly up to you. T-Rex skill runs the whole job. It finds the
winning format, breaks it down, makes the characters, picks the voices, writes the shoot plan, and then makes the video. One conversation, start to finish. Third, what else it can reach. Higsfield is a video and image studio.
Higsfield is a video and image studio. TEG is a catalog of over 4,000 tools. So the same agent that makes your video can also look up trends, research your competitors, and pull social data all through the same connection. Fourth, the
models. On the video side, they reach a lot of the same models. Seedance 2.5 is lot of the same models. Seedance 2.5 is on both. Tre also lists VO, WAN, Hiluo, and Miniax for video, plus Nano Banana Pro, GPT image, and Flux for images.
Fifth, open source. Tre's code is public. You can read exactly what it does, and a team can even run its own copy. And if you already have your own account with a provider, Tre uses your key first, so you stay in control.
Higsfield is a closed app, so you use it the way they built it. Now, where does Higsfield still win? If you never want to open Claude Code and you just want to click a button on a website, Higsfield is simpler. It has a polished interface,
lots of presets, and you can see everything visually. So, here's the honest way to think about it. If you like clicking through a website, Higsfield is great. If you already work with an AI agent and you want it to run
the whole video job for you, Tre is the one built for that. Let me show you the old way next to the new way. The old way to get a UGC ad looks like this. You scroll for hours looking for videos that are working. You message creators and
wait for replies. You send them your product and wait for it to arrive. You wait again for them to film it. Then you ask for changes. And when you finally get the video, you've got one version and no easy way to test anything else.
The new way looks like this. You tell your agent your niche and your product. It finds the format that's already winning. It makes your AI creator, your location, and your product shots. It picks the voice. It writes the plan.
makes a quick draft and then the final video. And you get three versions to test in one afternoon without leaving your desk. Now, let me show you how you'd actually use this in your business. If you run an online store,
start with your best-selling product. Say you sell a skincare serum. You ask your agent to find the most liked skincare videos from the last month and pick one where a creator shows a product to the camera. Then you ask it to make
your own AI creator holding your serum in a bright bathroom talking to the camera in a natural way. You ask for a little handheld wobble and natural window light so it feels like a phone video. You've got a product video that
looks like a real person filmed it ready for your ads. Say you sell hoodies. You ask for the street style format like the one in the demo. Two friends walking toward the camera showing off the fit. your AI creators, your hoodie, your
colors, and then three versions in three different places so you can see which one gets the most clicks. Say you're launching something new. Instead of waiting for creators to receive the product, your agent makes the launch
video the same day you have the product photos. That means your launch ads are ready before the stock even lands. And because the creator is locked, the same AI creator can appear in every ad for your store. So customers start to
recognize the face the same way they'd recognize a real person who's always in your videos. If you run an agency, this is a whole new service you can offer. Your client sends you their product and their best-selling ad. Your agent breaks
down the winning format, makes a creator for the brand, and makes a batch of videos. You hand the client three versions of every ad ready to test. You can also use it for your agency's own marketing. Make a short video for each
case study with an AI presenter walking through the result. And you can use the research step on its own. Before a client call, ask your agent to find the most liked posts in the client's niche and break down why they work. You walk
into the call knowing what's working in their market right now. If you make content, the same idea works for your own channel. Take one idea and ask your agent to turn it into a short video with your AI presenter in the format that's
already winning in your niche. One idea, one conversation, one finished video. Now, let me show you how to set this up. Step one, open Claude code or codeex.
Step two, paste this message into your agent. Read treg.2/lm.txt setup instructions, installs the T-reg command line tool, and asks you to sign
in. You can sign in with GitHub, Google, or a code sent to your email. Step three, paste this second message. Follow treg.2/skills/ai-c/skill.md
reference. Step four, answer the agent's questions. It'll ask about your niche, your product, your characters, and your voices, one question at a time. And here's the checkpoint. Before it makes any video, the agent shows you the plan
and asks you to approve it. So nothing gets made until you said yes. One tip, settle everything at the image stage first. If the face or the product looks wrong, fix it there before you ask for any video. That's the single biggest
thing that makes the final video look right the first time. Now you might be sitting there with a few doubts. The first one is I'm not technical. Claude code sounds scary. You don't write any code here. You paste two messages and
then you answer questions in plain English. The agent does the setup for you. The second one is, is it okay to use AI creators in my ads? Here's that rule I promised you. T-Rex's own playbook says remix the structure of a
playbook says remix the structure of a viral post, never the people in it. No real creator's face, no real creator's voice, no celebrity sound alikes. So you copy the format, not the person. And on top of that, label your AI videos as AI
and never present an AI creator as a real customer giving a real review. Show what your product does honestly. That keeps your ads trustworthy and keeps you safe on every platform. The third one is, will this work for my business? If
you sell anything you can show on camera or you run an agency that makes ads for people who do, this fits. Products, services, case studies, launches, the format comes from what's already winning in your niche. So, it's built around
your market, not someone else's. So, here's what you can do this week. Step one, pick [clears throat] your bestselling product or your best client. Step two, set up TEG in Claude code with those two messages. Step three, ask your
agent to find the most liked videos in your niche and break down the top one. Step four, make one AI creator, one location, and one video, and test it against your current best ad, one video at a time. And if you want help doing
that, join the AI profit boardroom. You get the AI video ad system with every message from this video ready to paste for stores and for agencies. You get the TE setup walkthrough plus the research, creator, voice, and shoot plan prompts,
so your agent runs the whole job for you. You get the full agent OS with the automation that plans, reviews, and builds your ads while you work on everything else. You get four live coaching calls every week where you can
bring your product and your ad and get it working. And you join over 3,400 business owners using these exact tools right now. The link is in the comments. and description or go to a profitboardroom.com.
So that's T-REG, an open-source Higsfield that lives inside Claude Code and Codeex. It finds a winning video format in your niche. It makes your AI creator and locks the face, the place, and the product. It picks a voice. It
makes the video on Seedance 2.5 with three versions to test. And one connection reaches over 4,000 tools so the same agent can research, plan, and build. Drop a comment and tell me, would you use this for your own store or for
your clients? I'll see you in the next one.
