OpenHiggsfield: A Unifying Studio for 40 Image and Video Models
A new open-source project landed today that consolidates the fragmented landscape of AI image and video generation into a single, clean interface. OpenHiggsfield — an open-source alternative to the paid Higgsfield AI — wraps 40 models (12 for images, 28 for video) behind one prompt bar, with per-model settings, a persistent gallery, and a self-hostable architecture.
In its first day on GitHub, it's already pulled 547 stars and climbing. The hosted version runs at openhiggsfield.ai with no install required — just bring a platform API key.
The model catalog is aggressive: Gemini Omni Flash, Kling 3 (Turbo/Std/Pro/4K/Motion), Veo 3.1, Wan, Flux, GPT Image 2, Ideogram, Recraft, LTX, MiniMax, PixVerse, Grok, Qwen, and a dozen more. Each model brings its own settings rail — aspect ratio, duration, batch size, output format — declared from a catalog file so adding a new model means one entry, not a UI rewrite.
Why this matters
The generation landscape is fragmenting. Every week brings another model with its own API, its own quirks, its own UI. What's missing is a composer — a tool that treats every model as an interchangeable engine behind the same workflow. OpenHiggsfield is that: a Zustand-powered Next.js 16 app where a prompt, a model pick, and a key are all you need.
Features that stand out:
- Media inputs by role — start frame, end frame, references — with per-role caps per model
- Live run lifecycle — skeletons on submit, polling every 4s, results bloom in as they finish
- Persistent history in IndexedDB with favorites, bulk actions, and 6-second undo
- Zero direct API exposure — browser talks only to server actions; the generation API key lives in an httpOnly cookie
- Self-hostable —
pnpm install && pnpm devgets you a local studio
The architecture is refreshingly clean. Five small Zustand stores, a catalog-driven model registry, and server actions as the sole API caller. It's the kind of project that makes you wonder why nobody built it sooner.
Not a replacement for any single model — but a replacement for the workflow of tab-hopping between five different generation UIs. That's a win.