OpenHiggsfield AI is a free, open-source alternative to Higgsfield AI that allows you to generate images and videos with 40+ AI models you prefer.
You can generate an image with one model, switch to a video model for the next step, and reuse an earlier result as source media. Prompts, model settings, and output links stay attached to each generation in the gallery. Reopen an earlier result to change the prompt, adjust a setting, or use the output in another generation.
OpenHiggsfield v.s. OpenHiggsfield AI
Higgsfield AI is built around a visual studio for working with image and video generation tools. OpenHiggsfield follows that idea with a public web app you can run on your own server.
OpenHiggsfield keeps image and video generation in a single workspace. Switch from Nano Banana to Kling or Veo, and the controls change to the options supported by the selected model. Previous runs stay in the gallery with their prompts and settings attached.
Supported Image & Video Generation Models
The composer changes with the selected model. Choose an image generator, and you can see controls such as aspect ratio, resolution, output format, prompt enhancement, or batch count when that model supports them. Switch to a video generator and the available fields can include duration, audio, start frame, end frame, reference media, or resolution.
Selecting a model changes the generation form to match the inputs and settings expected by that model. A Nano Banana image job can show image controls, while a Kling video job can expose duration, reference frames, audio, and other supported options.
| Model family | Models |
|---|---|
| Nano Banana | Nano Banana 2, Nano Banana 2 Lite, Nano Banana Pro |
| Kling | Kling 2.5, 2.6, 3 Turbo, 3 Std, 3 Pro, 3 4K, Motion variants, O1, O3 |
| Veo | Veo 3.1, Veo 3.1 Fast, Veo 3.1 Lite |
| Wan | Wan 2.6, 2.7, 3, 3 Prime |
| Flux | Flux 2, Flux 3 |
| LTX | LTX 2.5 Fast, LTX 2.5 Pro |
| MiniMax | MiniMax H3, Hailuo 2.3 |
| Other image models | GPT Image 2, Ideogram 4, Recraft 4.1, Qwen Image 3, Grok Imagine 2 |
Hosted Version and Self-Hosting
The hosted studio at OpenHiggsfield.ai asks for a generation-service key in this format:
id:secretThe key is stored in an httpOnly cookie with a 30-day maximum age. Production deployments mark the cookie as secure. Server actions read the credential when they submit a job or check its status.
To run the app on your own server, install the repository with pnpm:
pnpm install
pnpm devThe development server runs on port 3000. A deployment also needs these environment values:
HF_API_BASE_URL=
OPEN_HIGGSFIELD_READ_WRITE_TOKEN=How to Generate an Image or Video
Choose a model first, then enter your prompt and attach any reference media accepted by that model.
After you press Generate, OpenHiggsfield submits the request through a Next.js server action. A placeholder appears in the gallery while the job runs, and the app checks its status every four seconds.
Completed media replaces the placeholder. Failed, canceled, and NSFW responses are recorded with the returned reason. A request that does not reach a final state within ten minutes ends with a timeout in the studio.
One generation action can request up to four results. Models with their own count parameter use that field, while other models are submitted as multiple jobs and tracked individually.
Reuse Previous Results
Every finished generation keeps the prompt, model, resolved settings, timestamp, status, and output URLs in browser history. Opening a result lets you restore its model, prompt, and settings before running another version.
You can return to an earlier prompt, change only the aspect ratio or duration, or send a generated image into a video model that accepts image input.
The gallery has Image, Video, Assets, and Favorites views. It also supports multi-selection for downloads, favorite changes, and deletion. Shift-click extends a selection range, and deleted records can be restored for six seconds.
OpenHiggsfield stores up to 60 non-favorite generation records in IndexedDB. Favorites are kept beyond that limit.
Pros
- 40 image and video models.
- Hosted use and self-hosting.
- Model controls change with each generator.
- Previous prompts and settings can be restored.
- Generated media can be reused in later jobs.
- Batch generation supports up to four results.
Cons
- Generations require an external service key.
- No local inference.
- Uploaded media uses public Blob URLs.
Alternatives & Related Resources
- Open Generative AI: Free, Self-Hosted Image & Video Studio with 200+ Models
- OpenMontage: Open-Source AI Video Production System
- 10 Best AI Image Generators: Free and Paid
FAQs
Is OpenHiggsfield completely free to use?
The OpenHiggsfield studio does not currently require a subscription. Image and video generation uses your external service key, so model usage can still consume paid credits or incur charges from that provider.
Does self-hosting run the AI models locally?
No. Self-hosting runs the OpenHiggsfield web app on your server. Kling, Veo, Flux, Nano Banana, and the other supported models are accessed through the configured external generation service.
Are uploaded reference files private?
Not under the current upload implementation. Reference images, videos, and audio files are uploaded to Vercel Blob and receive public URLs before they are passed to the generation service.
