WTF AGENTS
Guide 25 / 37

WTF is Higgsfield

Fifty models, one chat agent, and now the code is open.

Fifty video and image models behind one chat agent, $700 million in annual revenue, a $5.4 billion valuation — and this week, an open API, eighteen months of open-sourced code and a $50,000 prize for whoever builds something with it first. What it is, what it costs, and why the open move matters.

The one-liner

Higgsfield is an AI video and image platform that puts fifty-plus generation models behind one interface and one credit balance, and increasingly behind one chat agent: describe the ad, the clip or the app you want, and Higgsfield chooses the models, runs them and hands back the result.

It is the tool most short-form creators and social media marketers actually use to make AI video, and in September 2026 it did something the closed labs have not: it opened its API to anyone, pay per generation, and a team member open-sourced a year and a half of the code that runs it.

By the numbers — September 2026

$5.4Bvaluation after the $400M Series B led by DST Global, 17 August 2026
$700Mannualised revenue at the time of the round, up from a $1.3B valuation in January
361Mvisits to the platform between January and July 2026
50+image and video models available through the new API, opened 16 September
18 monthsof the studio's code open-sourced by a Higgsfield engineer the same week
$50,000prize offered for practical builds on the API submitted within seven days
~$0.50cost of five seconds of Seedance 2.5 video through the API, per the open-source client's published rates

What it actually does

Higgsfield does not train its own frontier models. Its product is the layer on top: it licenses or integrates the best video and image models as they appear — Seedance, WAN, FLUX and dozens more — and wraps them in tools a creator can use without knowing which model is which.

Text and image to video — Short clips from a prompt, a photo or a product shot, with camera moves, presets and motion controls that were the company's first hit.

Character continuity — Face-Lock and its successors keep the same face, product or location consistent across shots, the problem that made early AI video useless for brands.

Genjutsu — Launched 1 September, on Product Hunt two days later: take the motion from any reference video and recast it with your own characters, locations and products.

Cinema Studio and Academy — A multi-shot editing workflow and a course library, aimed at people producing finished pieces rather than single clips.

Audio — Text-to-speech, voice change and lip-sync in the same credit balance, so a clip can leave with sound.

Supercomputer — The agent. Launched in May, it runs the whole pipeline from one chat: describe the piece, it picks the models, chains them and returns the result. Since summer it also builds and publishes whole generative apps — describe an app that makes images, video or 3D, and it wires the models in itself. No API keys, no code.

Similarity Scoring — For Team plans: flags likeness to known characters, celebrities and logos before delivery. A warning system, not legal clearance.

The pattern is the one running through this whole series: the models became commodities, and the value moved to the agent that chooses and chains them.

The September moves — API, open source, prize

Three things happened in one week, and together they change what Higgsfield is.

The API

On 16 September Higgsfield opened its API: over fifty image and video models, one pay-per-use balance in US dollars, separate from the consumer subscriptions. Any developer can now put Higgsfield's generation inside their own app, website or automated workflow. Until this week that layer was reachable only through Higgsfield's own interface.

The open source

The same week a Higgsfield engineer, Alisher S, open-sourced eighteen months of code — the creative-studio layer that sits on the API. A community client, Higgsfield-Open, followed within days: 38 of the models, exact per-generation pricing shown before you press go, and no platform markup. An independent project, OpenHiggsfield, had already been circulating as a free alternative with forty-plus models.

For a creator this means the interface is no longer the moat; the model access and the credit balance are. For Higgsfield it means the company is betting, as the Chinese labs did, that owning the ecosystem beats owning the front end.

The prize

Fifty thousand dollars for practical builds on the API submitted within seven days of launch. A small sum against a $5.4 billion valuation, and a loud signal about what the company wants next: developers, not just creators.

What it costs

RoutePriceNotes
Free$0Limited credits, watermarked, enough to learn the interface
Plus / Pro / UltimateMonthly, tieredCredits per month; higher tiers unlock more models, longer clips, higher resolution and "unlimited" windows on selected models
TeamCustomSeats, shared assets, Similarity Scoring, brand controls
APIPay per generationSeparate balance in USD; roughly $0.50 for five seconds of Seedance 2.5, about a cent for a Soul 2 image

Higgsfield's consumer pricing changes often and runs promotions constantly, so the page is the source of truth. The rule that does not change: everything is credits, video eats them fast, and an "unlimited" offer is always scoped to particular models at particular settings. Read the scope before you subscribe.

Higgsfield vs Runway vs HeyGen

HiggsfieldRunwayHeyGen
What it isMany models, one interface, one agentOne lab's own models and editorAvatar presenters and dubbing
Best forShort-form, ads, social, volumeCinematic control, VFXExplainers, sales video, localisation
Own modelsNo — integrates others'YesYes, for avatars
APIYes, since SeptemberYesYes
Open codeStudio layer open-sourcedNoNo
Price shapeCredits + pay-per-use APICreditsCredits

Runway is the studio for people whose video is the product. HeyGen is the presenter. Higgsfield is the assembly line for people who need a lot of good-enough video fast, which describes most marketing.

The catches

  • Credits burn. Video generation is expensive compute and every platform bills accordingly. Budget by finished clip, not by month, and set caps.
  • It rents its models. Higgsfield's strength is aggregation; its exposure is that any model it offers can be pulled, repriced or offered elsewhere. The API and open code are the answer to that, and a bet.
  • Rights are your problem. Similarity Scoring flags likeness; it does not license it. Generating a celebrity, a brand or a copyrighted character remains your legal risk, in every country you publish in.
  • Open-sourced is not open. The studio code is open; the models and the credit system are not. You still pay Higgsfield to generate.
  • Fast-moving means unstable. Genjutsu, Supercomputer's app builder, the API and the open-source drop all landed in four months. Features and prices in this guide will move; the shape will not.

Who it is for

Yes — Solo founders, creators and marketers who make short video weekly and want one bill, one interface and the newest models without tracking which lab shipped what.

Not yet — Anyone whose brand cannot risk a likeness or rights problem without legal review, or who needs frame-level control (that is Runway).

No — Anyone who just needs a talking presenter (HeyGen) or occasional clips (the free tier of anything).

Why it matters

Higgsfield is the clearest example in this series of a company winning by owning the layer between the models and the user. It trains nothing at the frontier and is worth $5.4 billion for choosing, chaining and packaging what others train. In September it took the next step: open the API, open the studio code, pay developers to build. That is the Chinese open-weight playbook applied to creative tools, and it is what an agentic economy looks like when the agent, not the model, is the product.

Glossary

Higgsfield
AI video and image platform aggregating 50+ models behind one interface, one credit balance and one agent. San Francisco; $5.4B valuation.
Supercomputer
Higgsfield's chat agent that runs the whole creative pipeline and builds generative apps from a description.
Genjutsu
Motion recasting: apply the movement of a reference video to your own characters and products. Launched September 2026.
Face-Lock
Character continuity across shots; the same face or product stays consistent.
Similarity Scoring
Team-plan tool that flags likeness to known characters, people and logos before delivery.
Higgsfield API
Pay-per-generation access to the platform's models, opened 16 September 2026.
Higgsfield-Open / OpenHiggsfield
Open-source clients for the API and a free alternative interface, respectively.
Seedance, WAN, FLUX
Third-party video and image models Higgsfield integrates. Names change quarterly.
Credits
The usage currency. Video consumes many; images few.
Aggregator
A platform that packages other companies' models rather than training its own. Higgsfield's model.

Liked this? Go deeper.

Higgsfield makes the video. Here is the stack around it, and the agents that run the business.

Get the rest of the series

Every WTF Agents guide is written the same way — plain English, no hype, no jargon. $7 each, or take a bundle.

Go deeper
Get the PDF guide · $7