Is fal.ai Down?

If Prismix shows fal.ai as "Operational" but you're having issues:

• Queue full / busy endpoint — fal.ai uses an asynchronous request queue for GPU jobs. During high-demand periods, newly submitted requests may sit in the queue for minutes before a GPU becomes available. Use the fal client's fal.queue.submit() method and poll the returned request ID rather than expecting a synchronous response.
• Webhook delivery failed — fal.ai can POST results to a webhook URL when a job completes. If your webhook endpoint is not publicly accessible (e.g. localhost during development), or returns a non-2xx status, fal will retry delivery with backoff and eventually stop. Use a tunnel like ngrok during local development and ensure your endpoint responds with 200 within 5 seconds.
• LoRA loading timeout on cold start — Custom LoRA adapters are fetched from storage at job start time. Large LoRA files (several hundred MB) can take tens of seconds to load on a cold GPU, causing the job to time out before inference begins. Keep LoRA checkpoints under 200 MB and prefer models that are already "warm" in fal's cache.
• Image generation 422 (invalid prompt or parameters) — fal.ai models reject requests that fail schema validation with HTTP 422. Common causes include mismatched image size parameters (width/height not divisible by 8 for diffusion models), unsupported negative prompt format, or model-specific required fields left empty. Check the model's schema in the fal dashboard.
• Credit exhausted mid-workflow — fal bills per-second of GPU time. If your account credit runs out while a job is in the queue or executing, the job fails with a payment error rather than an ordinary API error. Set a billing alert in the fal dashboard and top up credits before running bulk generation pipelines.
• Cold start latency on serverless endpoints — Serverless endpoints on fal spin down after inactivity to save cost. The first request after a cold period incurs a container startup + model load time of 10–60 seconds. For latency-sensitive applications, use fal's "dedicated" endpoint option which keeps a GPU warm, or send periodic keep-alive pings every few minutes.