6ioAPI
RUEN

fal models

The fal group opens the fal catalog: images, video, audio and other models. A model name is fal/<the model id at fal>, for example fal/fal-ai/flux/schnell. Generation runs as a task: the request returns an ID at once, with it you learn the status and collect the result when it is ready. fal calls this a queue, hence queue in the addresses. The addresses start with https://api.6io.io/native/fal, the authorization is the same: Authorization: Bearer sixio_sk_....

The body of the task goes to fal as it is. The fields of a model are on its page at fal.ai, the API tab: https://fal.ai/models/<model id>/api, for example fal-ai/flux/schnell.

A full example

import time
import requests

API = "https://api.6io.io/native/fal"
H = {"Authorization": "Bearer sixio_sk_..."}
MODEL = "fal-ai/flux/schnell"

# 1. the price of a unit (optional)
print(requests.get(f"{API}/price/{MODEL}", headers=H).json())

# 2. submit the task
job = requests.post(
    f"{API}/queue/{MODEL}",
    headers=H,
    json={"prompt": "a red circle on white background", "image_size": {"width": 1024, "height": 1024}},
).json()

# 3. wait until the status becomes COMPLETED
while True:
    st = requests.get(job["status_url"], headers=H).json()
    if st["status"] == "COMPLETED":
        break
    time.sleep(2)

# 4. collect the result
if st.get("error"):
    raise RuntimeError(st["error"])
result = requests.get(job["response_url"], headers=H).json()
print(result["images"][0]["url"])

Download the file at once or save the link: file links are valid for a limited time (see below).

Catalog and price

curl "https://api.6io.io/native/fal/models?q=flux&limit=5" -H "Authorization: Bearer sixio_sk_..."
curl "https://api.6io.io/native/fal/price/fal-ai/flux/schnell" -H "Authorization: Bearer sixio_sk_..."

The catalog is big, so it is paged: search by q and take the next page by next_cursor.

The answer about the price:

{
  "id": "fal/fal-ai/flux/schnell",
  "currency": "credits",
  "unit": "megapixels",
  "price_per_unit": "0.240000000000",
  "note": "The final charge depends on the billable units of the finished request (megapixels)"
}

unit is what the price is set for, price_per_unit is the price of one unit in credits (the markup is already included). Models with the units images, megapixels, seconds and videos are supported. For the others (for example, payment by compute time) submitting answers 400 unsupported_model. A model that fal does not have is 404 model_not_found.

Submitting a task

curl -X POST "https://api.6io.io/native/fal/queue/fal-ai/flux/schnell" \
  -H "Authorization: Bearer sixio_sk_..." \
  -H "Content-Type: application/json" \
  -d '{"prompt": "a red circle on white background", "image_size": {"width": 1024, "height": 1024}}'

The answer:

{
  "request_id": "req_8fK2xQ1mZpVwT9sJdL4nAg",
  "status": "IN_QUEUE",
  "status_url": "https://api.6io.io/native/fal/requests/req_8fK2xQ1mZpVwT9sJdL4nAg/status",
  "response_url": "https://api.6io.io/native/fal/requests/req_8fK2xQ1mZpVwT9sJdL4nAg",
  "cancel_url": "https://api.6io.io/native/fal/requests/req_8fK2xQ1mZpVwT9sJdL4nAg/cancel"
}

request_id and the links are ours, not fal's: the task belongs to your account, and nobody else's can be opened. Of the query string parameters only fal_webhook is accepted, the address for a notification about the end of the task; the others are not supported. The Idempotency-Key header works as in chat: see Retries and failures.

A notification instead of polling

Add ?fal_webhook=<address> (the address encoded) to the submit address, and we send the outcome to it when fal ends the task. The body of the notification is built like a fal webhook (request_id, status OK or ERROR, payload), but it is signed with our secret by Standard Webhooks, and the file links are ours. Details, signature verification and retries are on the Webhooks page.

Status, result, cancel

  • GET status_url: IN_QUEUE, IN_PROGRESS or COMPLETED (fal's words here, as in its SDK; in Sixio and jobs the states are queued, running, succeeded, failed). If fal ended the task with an error, the answer has an error field.
  • GET response_url: the model's result in the fal format. While the task is not ready, the answer is as at fal (400, "still in progress").
  • PUT cancel_url: cancel. 200 means the task was taken off the line and will not run, and the money is returned. 202 means the signal was sent to a running task: it may have reached the end, and then it is paid for.

File links in the result lead to our server and are valid for a limited time (14 days by default). Download what you need to keep longer.

Money

  1. Before submitting, an estimate is reserved: the price of a unit × the expected number of units (the number of images, the size in megapixels rounded up, the video length, 5 seconds by default).
  2. The final amount is charged by fal's data on the units actually billed. They appear a few minutes after completion. Until then the request is in the state billing_pending; this is normal, and you get the result at once.
  3. A task that fal ended with an error is not paid for: the reserve is returned after a short time.

Only a limited number of tasks can run at the same time. Images, video and audio have their own limit, separate from the chat limit. Details are in Errors and limits.