Higgsfield at $5.4 Billion: What the AI Platform Can Do and How Much Video Generation Costs

Higgsfield’s Business by the Numbers

Higgsfield primarily targets marketers, e-commerce, and advertising agencies. Business customers account for the bulk of the company’s revenue. Here’s what its 2026 growth looks like.

In January 2026, the startup raised $80 million as an extension of its Series A. The total round reached $130 million, and the valuation topped $1.3 billion. Investors included Accel, Menlo Ventures, and GFT Ventures.

In August, Higgsfield closed its Series B. At the time, annual revenue stood at $700 million, with an audience of more than 30 million users across 238 countries. The company said its customers included 390 Fortune 500 companies, and the number of generations exceeded 20 million per month. Higgsfield also highlighted growth in its agent product: in the three months after launching Supercomputer in May 2026, its user base grew 42x.

By the end of September, according to Bloomberg, the run rate had surpassed $1 billion. About 60% of revenue comes from subscriptions, and 40% from pay-as-you-go usage. As CEO Alex Mashrabov noted, companies that have achieved ROI from AI ads start producing more and more of them. Higgsfield is headquartered in San Francisco, with its engineering hub in Almaty.

The figures show that the platform’s core demand is driven by the advertising market, not individual creators. This explains why the most actively developed sections of Higgsfield are tied to marketing.

What the platform consists of

The Higgsfield interface is divided into several “studios.” Each one solves a specific task, but they all use a shared credit balance and a shared library of generations. 

Image and video generation. Core sections where you can pick a specific model and run a generation from a text prompt or a reference. Video models include Seedance 2.5 and 2.0, Kling 3.0, Wan 3.0, MiniMax H3, Sora 2, and Veo 3.1. For images, Nano Banana Pro, GPT Image 2.5, Seedream 5.0 Flash, the in-house Soul 2 model, and others are available.

Cinema Studio 4.0. A workspace for “AI filmmaking,” unveiled on August 12, 2026. The length of a single generation increased to 30 seconds, and the number of references to 50 images. The studio added an emotion wheel to control character performance, era selection with grain and color styling, four editing pace modes, and more than 30 camera movement presets. The Extend feature lets you continue an already finished video forward or backward.

Marketing Studio. An ad builder with a library of more than 1,500 templates. You feed it a product photo or a link to a product page, and it outputs product shots, banners, UGC clips, or motion videos. This section is covered in more detail in the practical part.

Genjutsu. A tool for reworking an existing video, launched on August 31, 2026. Motion Transfer transfers motion from the source clip to a new character, while Object Swap replaces a selected object, character, or product in the frame. Source video — 4 to 30 seconds, references — up to 30 images.

Supercomputer. A chat-based AI agent that plans the task on its own, selects models, and assembles the result. The agent retains project context between sessions, supports skills, connects to 30+ services like Slack, Google Drive, Notion, and Figma, and can run tasks on a schedule. Claude, GPT, and Gemini are available as the “brain” — on September 23, they added Claude Opus 5.5 to the list.

3D Jutsu. An AI 3D scene editor, released on September 4, 2026, alongside Supercomputer. It turns images into 3D models and lets you assemble scenes from them.

In addition to the web interface, Higgsfield is available via an MCP server for Claude, ChatGPT, and other agents, as well as via a ChatGPT plugin. The hands-on part of this article was done specifically via an MCP connection in Claude.

What Higgsfield API Offers

Until September, access to Higgsfield models from your own code was limited. With the launch of the open.higgsfield.ai platform, the API became a standalone product with pay-per-generation pricing and no subscription.

Authorization works via a key pair (ID and Secret) created in the developer console. Requests are processed asynchronously: the developer submits a job, then polls the status or receives a webhook when it completes. Finished files are stored for at least seven days, so you need to download the results to your side. SDKs are available for Python and TypeScript.

Here’s what a minimal request for generating a video via Seedance 2.0 looks like:

curl --request POST 
  --url https://api.higgsfield.ai/bytedance/seedance-2.0/text-to-video 
  --header "Authorization: Key HFAPIKEYID:{HF_API_KEY_SECRET}" 
  --header "Content-Type: application/json" 
  --data '{
    "prompt": "A cinematic tracking shot along a sunlit coastal road",
    "resolution": "720p",
    "generate_audio": true,
    "duration": 5,
    "aspect_ratio": "16:9"
  }'

API prices are listed in dollars per second of video or per image. Below is the cost of some of the models, according to open.higgsfield.ai. 

Separately, the company launched a cashback program for API generations. 

How much does Higgsfield cost

The web version runs on a subscription with a monthly credit bundle. Credits are deducted for each generation, and the cost depends on the model, duration, and resolution. 

On the Plus and Ultra plans, some models are available in Unlimited mode — with no credit deductions for seven days. For example, this includes Kling 3.0 and Nano Banana 2. However, unlimited and free generations work only on the higgsfield.ai website: they are not available via MCP, CLI, Canvas, or Supercomputer. For businesses, there are separate plans with team workspaces.

For reference: generating Nano Banana Pro costs 2 credits, a five-second Kling 3.0 video in 720p costs about 7 credits, and Seedance 2.0 in 720p costs about 22 credits.

Practice: an ad campaign for Incrypted merch

For the hands-on part, we used the scenario that Higgsfield promotes as the main one — advertising a product from a single photo. The product in this case is a hoodie with the Incrypted logo.

The work was done via Higgsfield’s MCP connection in Claude. All generations appear in the account library on higgsfield.ai, and the same steps are available in the Marketing Studio web interface. The Plus plan was used.

Step 1. Preparing the product photo

Marketing Studio accepts a product photo or a link to a product page as input. First, you need to create a “studio” photo of it. The Incrypted logo was provided to the model as a reference.

Prompt

E-commerce studio packshot of a premium heavyweight black oversized hoodie, merch of the media brand INCRYPTED. On the chest: the exact round red logo with the white key from the reference image, and below it a clean white uppercase wordmark INCRYPTED in a bold geometric sans-serif. Hoodie laid flat-ghost-mannequin style, front view, soft studio light, seamless light-grey background, visible fabric texture, crisp embroidery detail, no other text.

The model transferred the logo without distortion and added the wordmark with no spelling mistakes. The generation took less than a minute and cost 0.25 credits. 

Step 2. Product Shots and Banner

The next stage is a set of images for the product page and social media. For this, the Marketing Studio Image model is used: it takes a hoodie photo and a scene description. All three shots were run in parallel at 2K resolution with a 4:5 aspect ratio. 

First shot — studio shoot

Studio product shot: the INCRYPTED hoodie on a minimal dark concrete plinth, dramatic rim light, premium streetwear catalog look.

Second shot — a lifestyle scene with a model

Lifestyle shot: a young man wearing the INCRYPTED hoodie working on a laptop with trading charts in a cozy night coffee shop, warm neon reflections, candid editorial photo, logo clearly visible.

Third shot — an ad banner with a headline

Instagram ad banner for the INCRYPTED hoodie: model wearing the hoodie on a city rooftop at dusk, bold headline text WEAR THE KEY and small caption Incrypted merch dropu, clean ad layout.

In all three frames, the logo and lettering were preserved without distortion, and the banner text was generated without errors. In the studio shot, the hoodie is “floating” above the pedestal rather than lying on it — that works for conceptual advertising, but for a product card it should be regenerated. Each image cost 2 credits.

Step 3. UGC video

UGC is a format that mimics a real customer video shot on a smartphone. In Marketing Studio, these clips run up to 15 seconds and are generated with voiceover. For generation, the Marketing Studio model was used in UGC mode: vertical 9:16 video, 15 seconds, 720p. The prompt specifies the script and the character’s line.

Prompt

UGC-style selfie video: a friendly young creator in his apartment unboxes and puts on the black INCRYPTED hoodie, shows the red key logo to the phone camera and says in English: Okay, the Incrypted merch finally arrived. Heavyweight cotton, embroidered key logo, and it's actually warm. If you read crypto news every morning like me, this is your uniform. Natural handheld phone footage, casual energy.

The render took about 10 minutes. In the clip, the character opens the box, takes out the hoodie, shows it to the camera, and then appears wearing it. The line is voiced in full and without changes, and the lip-sync matches the speech. The logo and lettering are preserved in every frame. The clip cost 75 credits.

Step 4. Template-based motion video

In addition to freeform generation, Marketing Studio offers a template gallery. The Motion category includes Hypermotion, 2D Motion, Mixed Media, and SaaS Motion formats. For the test, we chose the Monospace Callouts template — a dynamic product presentation with text callouts.

Prompt

Black INCRYPTED hoodie with the embroidered red key logo. Callouts: HEAVYWEIGHT COTTON, EMBROIDERED KEY LOGO, OVERSIZED FIT. End card: INCRYPTED MERCH.

Total campaign cost 

The full set of assets — product photo, three product shots, a UGC clip, and a motion video — cost about 171 credits. On the Plus plan, that’s around 17% of the monthly bundle.

Most of the budget goes to video, while images cost just a few credits. Before each generation, Higgsfield shows the exact cost, so you can keep spending under control in advance. 

Higgsfield Limitations

Despite its broad toolset, the platform has a number of limitations worth considering before buying a subscription.

1. Video length in Marketing Studio is up to 15 seconds per generation. For longer ads, you need to stitch multiple clips together or use Cinema Studio.
2. Editing the result is available only in the Posters category. Other formats will have to be regenerated from scratch.
3. Voiceover and pacing in UGC videos may vary from one generation to the next.
4. Unlimited generations work only in the web interface — via MCP, API, and Supercomputer, credits are always deducted.
5. Video is the most expensive part of the workflow: a 15-second clip costs dozens of credits, and the Starter plan includes only 200 credits per month.
6. Video generation takes up to 10 minutes or longer, especially during peak hours.
7. Responsibility for the rights to any faces, logos, and products used lies with the user.

Most of these limitations are typical of the AI video market as a whole, not just Higgsfield. Still, before launching serial ad production, it’s worth testing the formats you need on a small budget.

Who Higgsfield Is For

Higgsfield is not a standalone model, but a layer on top of dozens of generators with ready-made business сценарии. The platform’s core audience is marketers, e-commerce teams, and ad agencies that need to ship a lot of visual content fast. A practical test showed that, from a single photo, you can produce a set of product shots, a banner, and two ad videos in just a few steps.

For developers, the September API launch provides access to the same models with pay-per-generation pricing. For creators who prefer working in chat, there is an MCP connection and the Supercomputer agent. On September 21, OpenAI published a case study on how Higgsfield ships new features in one day using GPT-6 Astra — the company continues to bet on agent-based ad production workflows.