What is the difference between Flux Dev and Schnell?

Flux.1 Schnell is distilled for speed — it produces good images in just 1 to 4 steps and costs less. Flux.1 Dev trades speed for higher fidelity and prompt adherence, using more steps and guidance. Pick Schnell for drafts, Dev for finals.

Does this tool generate images?

No. It builds the prompt and the request payload for fal.ai's API. You run the actual generation through your own fal.ai client or an HTTP call, since fal.ai requests are authenticated and best made server-side.

How many steps should I use?

Schnell-class models need only 1 to 4 steps. Dev and SDXL typically look best between 25 and 40 steps; beyond that you see diminishing returns and higher cost. The builder keeps your steps within each model's sensible range.

How is fal.ai priced?

fal.ai bills per generation or per compute-second depending on the model, with faster distilled models costing less per image. The estimate here is approximate; check fal.ai's current pricing for exact figures, especially at larger sizes.

fal.ai Model Prompt Builder

fal.ai model prompt builder

fal.ai hosts a catalog of fast, serverless image models — Flux.1 Dev and Schnell, SDXL, and Fast-SDXL — each with its own ideal parameter ranges. This builder writes a clean prompt and assembles the API input payload with the right image size, steps, and guidance for the model you pick, plus an estimated cost, so your first call lands instead of erroring on a bad parameter.

How it works

You choose a model and the builder adapts: Schnell-class models cap inference steps low (1–4) because they are distilled for speed, while Dev and SDXL open up to 25–40 steps for fidelity. It then emits a JSON input object — prompt, image size, num_inference_steps, guidance_scale — alongside the fal.ai endpoint ID, ready to paste into a fal.ai client or an HTTP request. Generation itself runs on your side because fal.ai calls are authenticated and best kept server-side.

Tips and notes

Draft on Schnell, finish on Dev. Iterate prompts cheaply at 2–4 steps, then re-run the winner on Flux.1 Dev with more steps.
Guidance is model-specific. Flux Dev likes a low guidance scale (around 3.5); SDXL prefers higher (7–9). The builder seeds sensible defaults.
Match size to use. Larger images cost more and take longer — generate at the size you actually need.
Keep generation server-side. Drop the payload into a backend call; fal.ai keys should not ship to the browser.