Video Generation Media Platform Higgsfield Launches Unified API for Over 50 Models, Pay-Per-Use Pricing Without Subscription Binding
The video generation media platform Higgsfield has launched a developer interface, claiming that a single asynchronous authentication interface can call over 50 cutting-edge image and video models, with pay-per-use pricing in USD and no subscription binding, at a lower price than subscription packages.
The catalog includes models from Seedance, Kling, Wan, MiniMax, LTX, PixVerse, Recraft, Ideogram, Grok Imagine, as well as its own models Soul 2, Soul Cinema, DoP, and Marketing Studio Image. The process is unified into submission for generation, polling or webhook, and then downloading; Python and TypeScript SDKs and REST are provided. After creating a key, the default concurrency is 20.
Billing is separate from the official website subscription: pre-charge USD balance, with a minimum recharge of 5 USD; if the balance is zero, new requests will queue. Images are billed per image, videos per second, and DoP per use. Public starting prices include approximately 0.0032 USD/image for Soul 2, 0.042 USD/second for Kling 2.5, 0.112 USD/second for Kling 3.0, 0.20 USD/second for Wan 3.0, and 0.9332 USD/second for Seedance 2.0.
Promotions offer a default 15% discount on all models, and users can select up to three to four commonly used models to increase the discount to 50%, which must be locked within seven days; verifying a corporate email and adding approximately 15 USD to the balance can unlock the first key. On Product Hunt, it launched with about 148 votes, ranking 6th. The company is also promoting it in conjunction with GPT-6 Astra, targeting developers who want to integrate generation into their own products rather than web creator packages.
The official Unlimited and MCP channels still operate on a subscription and quota basis; the API is a separate account. The catalog is intentionally narrowed to the "most commonly used versions in the current production pipeline," with each model page listing resolution, duration, sample code, and comparison tables.
In market mechanisms, this is an aggregation distribution interface layer: what is bought is the labor hours of connecting to five suppliers less, outsourcing the queue and GPU; what is sold is the directory fragmentation and seat fees of each official API. Beneficiaries are application parties that need to mix Kling and Seedance in their products; under pressure are web generation platforms that rely on subscription locks for usage, and small to medium teams that must maintain multiple model gateways. The discount window is used to lock in the unit prices of the three main models into the counterpart's cost table.
Additionally, the documentation splits Seedance 2.0 into three endpoints: text-to-video, image-to-video, and reference-to-video, returning request_id and status_url; automatic recharge has been removed from the console, showing daily and model-based costs.
Source: Public Information
ABAB AI Insight
Higgsfield initially retains creators on-site through a web generator and Unlimited subscription, then sells the same batch of third-party models as pay-per-use APIs to developers. This shifts from "human pressing buttons" to "programs adjusting queues." Proprietary models like Soul and DoP are used to anchor lower prices, while Kling and Seedance demonstrate that the catalog is not an empty shell. Compared to Replicate, Fal, and OpenRouter, the difference is that it markets itself as "cheaper than subscriptions" and uses a seven-day 50% discount to lock in the switching costs of the three main models.
Capital is flowing not to self-training bases, but to gateways, billing, and concurrency. Pre-charged balances, 20-way concurrency, and webhooks turn GPU rental and queuing into billable intermediaries. The 50% discount only locks in three to four models, indicating that the real target is those few high-frequency endpoints in the production pipeline, not long-tail experiments. Tying it with GPT-6 Astra connects the entry for agents writing code to the generation output, shortening the path from "prompt to embeddable product."
In contrast to Stability's initial distribution of open-source weights, Runway's integration of models into creative suites, and OpenAI's embedding of video capabilities into its own products: aggregation layers always appear after models are scattered. The industry phase is characterized by model surplus and integration scarcity. Whoever can consolidate 50 endpoints into a single authentication will capture the access budget of small to medium applications first. Web subscriptions will not die immediately, but growth will shift towards pay-per-use interfaces.
The structure belongs to the upward shift of distribution rights in technological substitution. Model companies sell seconds and images, while application companies do not want to maintain five sets of keys; aggregators use discounts and unified lifecycles to shift pricing power from "model official package" to "gateway directory." The mechanism is: generation costs can now be quoted per second, while integration costs are still calculated per engineer week; when the latter is more expensive, the market will pay for the intermediary layer, even if the intermediary does not own the models.
ABAB News · Cognitive Laws
The more models there are, the more valuable a unified interface becomes.
Subscriptions lock users in, while pay-per-use turns users into a flow.
The discounts do not lock in hobbies, but the three models in the pipeline.