Skip to content
Zabio is in beta: models, prices and features can change. Tell us what breaks at hello@zabio.ai.

Videos

Zabio makes videos through POST /v1/videos, in the shape OpenAI’s SDKs knew for Sora. A video takes one to several minutes to make, so the request answers at once with a video id; you fetch the video when it is ready. Change base_url and your key, keep the rest.

For the exact shape of the request and the response, see the API reference.

Send a prompt, a length and a size. The key needs the chat scope (keys have it by default).

Terminal window
curl https://api.zabio.ai/v1/videos \
-H "Authorization: Bearer insert your zabio.ai key here" \
-H "Content-Type: application/json" \
-d '{"model": "auto", "prompt": "A red lighthouse on a rocky shore at dusk, gentle waves, slow camera pan", "seconds": "5", "size": "1280x720"}'

The answer comes at once, with status 202:

{"id": "video_...", "object": "video", "status": "queued", "model": "auto", "seconds": "5", "size": "720p", "progress": 0, "created_at": 1793743200, "expires_at": 1796335200}

Then ask for it with GET /v1/videos/{id} every few seconds. While it is made the status is queued or in_progress, with progress from 0 to 100 where the model reports it. Once it is completed:

{
"id": "video_...", "object": "video", "status": "completed", "model": "auto", "seconds": "5", "size": "720p", "progress": 100,
"created_at": 1793743200, "completed_at": 1793743390, "expires_at": 1796335200,
"zabio": {"model_served": "grok-imagine-video-1.5-lite", "vendor": "xai", "cost_micros": 105000, "call_id": "call_...", "file_id": "file_...",
"url": "https://api.zabio.ai/v1/videos/content/file_...?exp=1793746990&sig=...", "url_expires_at": 1793746990}
}

Download the MP4 with your key at GET /v1/videos/{id}/content, or through zabio.url, a link that works without the key for 60 minutes from the request that returned it, so it can go straight into a <video> tag. Anyone holding the link can open the video until then: share it as you would the video. Ask for the video again to get a fresh link. The video is also a file: GET /v1/files/{file_id}/content returns it.

Model Lengths Resolutions Starts from Sound Price per second
grok-imagine-video-1.5 (xAI) 1 to 15 s 480p, 720p, 1080p a prompt, an image, a first and a last frame on, can be turned off 0.08 USD, plus 0.01 per image sent
grok-imagine-video-1.5-lite (xAI) 1 to 15 s 480p, 720p, 1080p (upscaled from 720p) a prompt, an image on, can be turned off 0.02 USD, plus 0.01 per image sent
wan2.6-t2v (Alibaba) 2 to 15 s 720p, 1080p a prompt always 0.10 USD at 720p, 0.15 at 1080p
wan2.6-i2v (Alibaba) 2 to 15 s 720p, 1080p an image always 0.10 USD at 720p, 0.15 at 1080p
gemini-omni-1.1-flash (Google) 3 to 10 s 360p, 720p, 1080p and 4k (the last two upscaled) a prompt, an image, a first and a last frame always 17.50 USD per million video tokens, about 0.10 USD a second at 720p

Prices are the vendors’ list prices; the plan fee applies on top. GET /v1/models lists each video model’s lengths (zabio.media.durations), resolutions (zabio.media.sizes), inputs (zabio.media.inputs), sound (zabio.media.audio) and price (zabio.pricing.per_second, or per_second_by_resolution). The xAI models are made on xAI’s global endpoint, so their processing location is not guaranteed; the Alibaba models are made in the United States (Virginia); Google’s model in the United States.

gemini-omni-1.1-flash makes the video while Zabio waits on Google’s side, so it can take longer than the other models before the status turns completed. It makes 3 to 10 seconds, the length set by seconds as on the other models (Google’s API has no length field, so Zabio asks for it in the prompt it sends, and the model makes it: 3 s asked was 3 s made when Zabio tested it on 2026-11-05). It makes landscape (16:9, the default) or portrait (9:16) videos; a square size gets landscape. It is billed on the tokens Google reports: video tokens at 17.50 USD per million, and the resolution changes how many a second costs (about 1,931 a second at 360p when Zabio tested it, about 0.034 USD a second; 5,792 at 720p as Google states, about 0.10 USD a second; no rate stated for 1080p and 4k, which are upscaled), the prompt and the images at 1.50 USD per million input tokens, and the model’s thinking at 9.00 USD per million.

Google’s stated limits apply: Gemini Omni does not make videos showing certain recognizable people, from a prompt or from an image, and such a request fails with moderation_blocked and a message that says so. Editing images of minors, and editing or extending uploaded videos, are not available in the European Economic Area, Switzerland and the United Kingdom; Zabio takes no uploaded video for generation, so the second limit does not reach your requests today.

model: auto (the default) picks the cheapest model that makes the length and the resolution you ask for, from what you send (a prompt, an image, a last frame), with the sound you ask for, by the vendors’ own prices. Video has no quality bar yet, so every auto line picks the same model. A length or resolution no model makes is a 400 that names each model’s limit.

Field What it does
model a video model, or auto
prompt what the video shows, up to 4,000 characters. Optional when you send an image to a model that starts from an image alone
seconds the length in whole seconds, as a string ("5") or a number. Default: 8 seconds where the model makes 8, else its shortest
size WIDTHxHEIGHT, read as its resolution (the smaller side: 480 is 480p, 720 is 720p, 1080 is 1080p) and its orientation (landscape, portrait or square). The model makes its own pixels at that resolution, so the answer’s size is the resolution
resolution instead of size: 480p, 720p, 1080p or 4k. Default: the model’s lowest
input_reference an image to start from (the first frame): a base64 data URL of a png, jpeg or webp, {"image_url": "data:..."}, {"file_id": "file_..."} of a file you uploaded with purpose user_data, or a file in the multipart form the SDKs send. At most 20 MB. Web addresses are not fetched
zabio.audio false for a silent video, on a model whose sound can be turned off
zabio.last_frame the last frame, in the forms of input_reference, on a model that takes two images
zabio.expires_after {"anchor": "created_at", "seconds": N}, N at least 3600: delete the video sooner than your organization’s retention

One video a request. The zabio object is a Zabio extension; the OpenAI SDKs pass it through extra_body, and a multipart form carries it as one field named zabio holding the JSON. remix, thumbnails and webhooks are not available: ask for the video’s status instead.

The video is made on Zabio’s side. Zabio asks the model’s vendor to start it, asks how it is going every few seconds, and as soon as it is ready fetches it, stores it and bills it. The vendors keep their copy for a day at most; Zabio’s copy is the one you download. A video that cannot be made ends failed with an error that has a code and a message. A video Zabio could not keep after the vendor made it bills nothing; one that was billed and then lost has its charge returned to your balance, and its error says so. A video still not ready six hours after it started ends failed with generation_timeout and bills nothing.

DELETE /v1/videos/{id} cancels a video that is still queued, and nothing is billed. One already being made answers 409 video_in_progress: it is at the vendor and will be billed. Once it has ended, DELETE removes it and its file.

GET /v1/videos lists your videos, newest first, as the status objects above (without a link: ask for one video by its id for that). limit (1 to 100, default 20) sets the page size, after takes the last_id of the previous page, and status keeps one status. A key in a project sees its project’s videos.

Each second of the video made, at the model’s price for the resolution made: the seconds the vendor bills (Alibaba reports them; xAI reports the video’s length), plus the model’s price per reference image where it has one. On gemini-omni-1.1-flash, the tokens Google reports (see Gemini Omni above). The plan fee applies on top, and the auto fee when auto picked the model. A video that failed bills nothing, except where the vendor charges for a refusal: xAI charges for a video its moderation refuses (0.02 USD on grok-imagine-video-1.5-lite when Zabio tested it on 2026-11-05), and then the request is billed what xAI charged plus the plan fee, and its error message says how much.

Credit is checked when the video is queued. The video’s estimated cost (the length times the model’s highest price per second, fee included) is set aside on your balance, so videos queued at the same time cannot take the balance below zero together. A video whose estimate does not fit the balance left after the calls still in flight is refused with 402 insufficient_credits: the message says the estimate, the balance and the least amount to add, and nothing is queued. The amount set aside is given back when the video fails or is cancelled, and is replaced by the real charge when the video is made. Credit is checked again when the video starts: one whose balance reached zero while it waited ends failed with insufficient_credits and bills nothing. The free plan does not make videos.

The prompt and the images you send go to the vendor that makes the video. Zabio stores the request (the prompt and the images included) until the video ends, and the video until you delete it or your organization’s retention runs out (30 days by default), or sooner with zabio.expires_after; both are encrypted with your organization’s key. With per-key body logging on, the prompt and a reference to the images and the video (count, type, bytes, length) are kept under the retention, never their bytes. Tell your users that the videos are AI-generated where you show them, as the vendors’ terms require.

Status code When
400 invalid_value a field is out of range (seconds not a whole number, a size that is not WIDTHxHEIGHT, a resolution outside the list, size and resolution that disagree, n above 1, a prompt above 4,000 characters, zabio.expires_after under 3600 seconds), or a named model does not make the seconds or the resolution; param says which
400 invalid_value no prompt and no input_reference; a zabio.last_frame without input_reference; an input_reference that is not a png, jpeg or webp, or a file_id your key cannot read
400 unsupported_value no model can make the video asked; or the named model does not start from an image (or needs one), takes no last frame, or always makes sound (zabio.audio: false); a web address as input_reference; variant other than video on /content
400 unsupported_for_model the named model is not a video model
402 insufficient_credits the video’s estimated cost, fee included, does not fit the balance left after the calls still in flight, each of which has its estimate set aside until it ends (ask for a shorter or lower-resolution video, or add credits); also the error of a video whose balance reached zero before it started
404 model_not_found not a model, or not allowed for the key
404 video_not_ready /content before the video is completed
404 not_found no video with that id; a link that expired, was changed, or whose video was deleted
409 video_in_progress DELETE on a video being made
410 expired /content of a video that was deleted or expired
410 model_retired the model is past its retirement date (the message says the date); nothing reached the vendor
413 media_too_large, payload_too_large an image above 20 MB, or a body above about 55 MB
503 video_unavailable, no_provider_available video generation is not available on this server, or the model’s provider is not configured

A failed video’s error.code is one of: moderation_blocked (the vendor’s safety system refused the prompt, the image or the video), invalid_value (the vendor rejected an input), generation_interrupted (the server restarted at the wrong moment; the message says whether anything was billed), generation_timeout, insufficient_credits, storage_unavailable, internal_error (an error on Zabio’s side; when the video was already billed the charge is returned and the message says so), no_provider_available, api_key_revoked, key_expired, or the vendor failure codes of the errors page. The key’s requests per minute limit applies. A video that was already made, stored and billed is yours even if Zabio hits an error while finishing it: it ends completed and you can download it.

import time
from openai import OpenAI
client = OpenAI(base_url="https://api.zabio.ai/v1", api_key="insert your zabio.ai key here")
video = client.videos.create(model="auto", prompt="A red lighthouse on a rocky shore at dusk, slow camera pan", seconds="5", size="1280x720")
while video.status in ("queued", "in_progress"):
time.sleep(10)
video = client.videos.retrieve(video.id)
if video.status == "completed":
client.videos.download_content(video.id).write_to_file("lighthouse.mp4")
print(video.status, video.model_extra.get("zabio"))
# From an image, silent
video = client.videos.create(model="grok-imagine-video-1.5-lite", prompt="The waves move", seconds="4",
input_reference=open("frame.png", "rb"), extra_body={"zabio": '{"audio": false}'})

With the multipart form the SDKs send when there is a file, the zabio field is a JSON string, as above; on a JSON request it is an object.