Media jobs

Images, video, music and speech are generated by jobs, because a render takes from seconds to several minutes.

Start

POST /v1/{images|videos|music|audio}/jobs with a model of that kind. The answer comes back at once with id, status: "pending" and estimated_cost.

Poll

GET /v1/{kind}/jobs/{id} every 3–5 seconds until status is completed or failed. The job keeps running if you stop polling.

Download

Each output entry has a url (…/v1/{kind}/jobs/{id}/content?index=N). It serves the file with the same X-API-KEY; Range requests are supported.

Kind	Render time	Retention
Images	Seconds to a couple of minutes (4K up to 5)	7 days
Video	30 seconds to a few minutes	7 days
Music	Several minutes	Download right away
Speech and sound	Depends on the model	Download right away

The exact deadline is in expires_at; after it you get 410 content_expired. An account can run a limited number of jobs at once (429 too_many_jobs).

Images

POST /v1/images/jobs — text-to-image, or editing when input_images are given. One image per job.

Field	Description
`model`, `prompt`	Required. Prompt usually up to 5000 characters
`size`	`1:1`, `16:9`, `1k`/`2k`/`4k`, `WIDTHxHEIGHT` or `auto`
`input_images`	References (https or `data:` URI), usually up to 10
`output_format`	`png` (default), `jpeg`, `webp`
`quality`	`standard`, `hd`
`mj_task_id`, `mj_index`	Midjourney actions

Video

POST /v1/videos/jobs. The mode follows the inputs: text only; image_url (+ last_frame_url) for image-to-video; reference_image_urls / video_urls for reference-to-video; video_url + image_url for motion control.

Field	Description
`model`, `prompt`	Required. Prompt up to 10,000 characters
`duration`	Seconds; each model has its own set (Veo 4/6/8, Kling 5/10, Seedance 4–15)
`aspect_ratio`	`16:9` (default), `9:16`, `1:1` and more
`resolution`	`480p`, `720p`, `1080p`, `4k`
`audio`	Sound track where supported (may raise the price)
`mode`	`standard`, `pro`, `master`
`output_format`	`mp4` (default), `mov`, `webm`

Music

POST /v1/music/jobs. Send only prompt and the model writes the lyrics and picks a style. Fields: lyrics (up to 5000), style (up to 1000), title (up to 200), instrumental. Some models return two tracks for one price.

Speech and sound

POST /v1/audio/jobs. Speech — text + voice; dialogue — dialogue (turns with text and voice); sound effects — text + duration_seconds (up to 30); voice isolation — audio_url. Optional: language_code, speed (0.7–1.2), stability, similarity_boost.

Example

import os, time, requests

BASE = "https://b2b.api.ascn.ai/api/ai-gateway"
HEADERS = {"X-API-KEY": os.environ["ASCN_API_KEY"]}

job = requests.post(f"{BASE}/v1/images/jobs", headers=HEADERS, json={
    "model": "google/nano-banana-2",
    "prompt": "A cozy cabin at sunrise, cinematic light",
    "size": "16:9",
}).json()

while job["status"] not in ("completed", "failed"):
    time.sleep(4)
    job = requests.get(f"{BASE}/v1/images/jobs/{job['id']}", headers=HEADERS).json()

print(job["output"][0]["url"] if job["status"] == "completed" else job["error"])

Job errors

`error.code`	What to do
`content_moderation`	Rephrase the prompt
`input_rejected`	Change the input file, see `message`
`invalid_request`	Fix the fields as `message` says
`generation_failed`, `job_expired`	Start a new job

A failed job is not charged. Reference files must be reachable over https without signing in.

Getting Started

Guides

API

Images

Video

Music

Speech and sound

Example

Job errors

​Images

​Video

​Music

​Speech and sound

​Example

​Job errors

Images

Video

Music

Speech and sound

Example

Job errors