Generating media

Images, speech and video. The endpoints follow the OpenAI shapes, so the SDK methods you already know work — you just name an Upfyn model.

Which model generates what#

Images#

1image = client.images.generate(
2    model="upfyn-yuva",
3    prompt="A wide shot of a rain-soaked Jaipur street at dusk, warm streetlights",
4    size="1024x1024",
5    n=1,
6)
7 
8print(image.data[0].url)

Editing an image

POST /v1/images/edits takes an uploaded image or an image_file_id you already own. One of the two is required. Passing a file id to the generations endpoint is rejected rather than ignored — edits and generation are separate routes on purpose.

Speech#

1audio = client.audio.speech.create(
2    model="upfyn-yuva",
3    voice="alloy",
4    input="Your order has shipped and arrives on Thursday.",
5    response_format="mp3",
6)
7 
8audio.stream_to_file("notice.mp3")

Video#

Video is always asynchronous — it returns a task you poll.

1curl https://ai.upfyn.com/v1/videos \
2  -H "Authorization: Bearer $UPFYN_API_KEY" \
3  -H "Content-Type: application/json" \
4  -d {
5    "model": "upfyn-rishi",
6    "prompt": "A paper plane crossing a sunlit office, slow motion",
7    "seconds": 6,
8    "size": "1280x720"
9  }

1{
2  "id": "task_8f21...",
3  "object": "video.generation.job",
4  "status": "queued",
5  "poll_url":   "https://ai.upfyn.com/v1/tasks/task_8f21...",
6  "events_url": "https://ai.upfyn.com/v1/tasks/task_8f21.../events",
7  "cancel_url": "https://ai.upfyn.com/v1/tasks/task_8f21.../cancel",
8  "credits_reserved": 75.62
9}

Collecting the result

• Poll poll_url — statuses are queued, running, submitted, retrying, succeeded, failed, cancelled.
• Or subscribe to events_url, an SSE stream, instead of polling in a loop.
• On success the task carries output[] with a signed url, an expires_at and a file_id.
• Asking for the content before it is ready returns 409 video_not_ready.
• cancel_url stops a job you no longer want and releases the reservation.

Making images and speech async too#

Image and speech calls run synchronously by default. For long jobs, ask for a task instead — any of these three works.

1curl https://ai.upfyn.com/v1/images/generations \
2  -H "Authorization: Bearer $UPFYN_API_KEY" \
3  -H "Prefer: respond-async" \
4  -H "Content-Type: application/json" \
5  -d '{ "model": "upfyn-yuva", "prompt": "A quiet library at night" }'

• The Prefer: respond-async header.
• "async": true in the body.
• "background": true in the body.

You get a 202 with the same task shape as video, plus a Location header and Retry-After.

Where the files go#

Everything generated is stored against your account and appears in Files.