Images and video

Images come back immediately, and video is made asynchronously: you create a job, wait for its status and download the file. Both APIs are compatible with the OpenAI SDK.

ImagesCopy Anchor Link

import base64
import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.morphogen.ru/v1",
    api_key=os.environ["MORPHOGEN_API_KEY"],
)
resp = client.images.generate(
    model="flux-2-klein",
    prompt="красное яблоко на белом столе",
    n=2,
)
for i, image in enumerate(resp.data):
    with open(f"apple-{i}.jpg", "wb") as f:
        f.write(base64.b64decode(image.b64_json))

The image arrives as a b64_json string. The response_format: "url" mode is not supported: there is no image hosting. Streaming is not supported either. Parameters: Images.

Image models also work through /v1/chat/completions with modalities: ["image"]: the image then arrives in choices[0].message.images[].image_url.url as a data URL. This is handy if your app is already built on chat.

VideoCopy Anchor Link

1. Create a job

   Copy Text

   video = client.videos.create(
       model="grok-imagine-video",
       prompt="бумажный кораблик на озере на закате",
       seconds="4",
       size="854x480",
   )
   print(video.id, video.status)

   The response comes back immediately with the queued status.

2. Wait for the result

   Copy Text

   import time
   
   while video.status in ("queued", "in_progress"):
       time.sleep(10)
       video = client.videos.retrieve(video.id)
   print(video.status)

   A clip usually takes up to a couple of minutes to generate. The failed status means generation failed, and no money is charged for it.

3. Download the file

   Copy Text

   client.videos.download_content(video.id).write_to_file("boat.mp4")

4. Delete it when you no longer need the file

   Copy Text

   client.videos.delete(video.id)

   Until you delete the job, the file is stored on our side.

The same calls are available over plain HTTP, and the SDK is optional: Create, Get, Download, Delete.

MoneyCopy Anchor Link

Before a video starts, an amount is reserved: the seconds multiplied by the price per second at the chosen resolution. If you do not set generate_audio, a model with sound takes the price with sound. After completion, the actual cost is charged. A job cancelled in the queue and any job with the failed status cost nothing. For images, the reserve equals the number of images multiplied by the price of one.

ErrorsCopy Anchor Link

All codes: Errors.