Learn fal
Fast hosted inference for image and video models with a playground per model and an API. The quick way to try Flux, Kling, Wan and others.
What fal is good at
- Trying new models the week they appear
- Image-to-video across several models in one place
- Fast generation for iteration
- API access for pipelines
Quickstart
- Pick the modelThe gallery lists image, video, audio and utility models; open one to see its inputs.
- Use the playgroundFill the form, run, and inspect the result and the request details.
- Fix the seedOne change at a time.
- Keep the requestThe request page is the record of inputs and output; save its link and the output file.
Settings cheat-sheet
| Setting | What it does | Start with |
|---|---|---|
| Model | Which model | By task, then by samples |
| Image input | Start frame or reference | Clean, correct ratio |
| Prompt | Text input | Literal |
| Duration / size | Clip length or image size | Shortest or smallest first |
| Seed | Repeatability | Fixed while tuning |
| Safety / other | Model-specific toggles | Defaults |
fal guides
Guides for fal are on their way. The quickstart and cheat-sheet above are kept current in the meantime.
The five most-asked questions
What is fal?
A hosted inference platform for fast image, video and audio models with a web playground per model and an API, paid per use.
Which models are on fal?
Flux variants, several video models (Kling, Hailuo, Wan, LTX and others at various times) and utilities such as upscalers. The catalogue changes often.
How does fal differ from Replicate?
Both host open and partner models behind APIs and playgrounds. fal emphasises speed and video models; Replicate has a broader catalogue and detailed prediction pages. Many creatives use both.
How is usage billed?
Per run, by model and output size or duration; check the model page and pricing page.
Can I use outputs commercially?
Each model's licence applies and is linked on its page; record the model name and version with every output.