A text-to-video API: a six-second clip from a prompt
Add AI video generation to your app with one API key using asynchronous jobs, generating 6-second clips with sound for $0.54 per render.
On this page 8 sections
The most direct way to add AI video generation to your app is through an asynchronous jobs API that manages rendering and stores the resulting clip automatically. thelastapi provides this with vibiz/video, which runs LTX-2.5 Fast to generate six-second clips at 720p with sound from a text prompt or an existing still image. You authenticate with one key, poll a single job status endpoint, and receive a permanent media link stored directly in your project.
This guide walks through submitting video jobs, animating existing images, choosing aspect ratios, and managing budgets inside your application.
How video generation differs from chat completions
Text and voice generation return data within seconds or stream tokens over a server-sent events connection. Video rendering takes longer because the model must synthesize dozens of coherent frames and synchronize audio tracks.
Because HTTP requests will time out during long render times, video generation runs as an asynchronous job:
- Job creation: Your app posts the prompt, aspect ratio, and optional source image to
/v1/ai/jobs. - Immediate response: The API answers with an HTTP
202 Acceptedstatus and a unique job identifier. - Status polling: Your app polls
/v1/ai/jobs/:idevery few seconds to check whether the status is pending, processing, or completed. - File delivery: Once completed, the response returns the final video link, metadata, and exact cost. When run within a project, the file is saved permanently to Files.
This workflow prevents hanging sockets on your servers and allows background workers to handle video creation cleanly.
Submitting a video job with code
To generate a video, send a JSON payload specifying the job type as video and the model as vibiz/video. You can test this immediately from your terminal using the CLI or via HTTP requests.
Using the CLI
Run vibiz video with your prompt to generate and download a clip in one step:
vibiz video "A golden retriever running through autumn leaves in a park" \
-o dog.mp4 \
-aspect 16:9Using HTTP and curl
To integrate video creation into backend services written in Python, Go, or Ruby, call the AI jobs endpoint directly:
curl -X POST https://thelastapi.dev/v1/ai/jobs \
-H "Authorization: Bearer $VIBIZ_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"type": "video",
"model": "vibiz/video",
"input": {
"prompt": "Ocean waves crashing against basalt cliffs at dawn, cinematic lighting",
"aspect_ratio": "16:9"
}
}'The gateway returns a 202 Accepted response with the created job:
{
"id": "job_01k8v92m7x4n",
"object": "job",
"type": "video",
"status": "pending",
"model": "vibiz/video",
"created_at": "2026-10-28T12:00:00.000Z"
}Polling for completion
Poll the job status endpoint until the render completes:
curl https://thelastapi.dev/v1/ai/jobs/job_01k8v92m7x4n \
-H "Authorization: Bearer $VIBIZ_API_KEY"When the render finishes, the status changes to completed and returns the file details:
{
"id": "job_01k8v92m7x4n",
"object": "job",
"type": "video",
"status": "completed",
"model": "vibiz/video",
"served_model": "ltx-2.5-fast",
"cost_micro_usd": 540000,
"output": {
"data": [
{
"url": "https://thelastapi.dev/f/Qm9x8172938475610293847561029384/2026-10-28-ocean-waves.mp4",
"file_id": "file_01k8v93b9x2p",
"file_url": "https://thelastapi.dev/f/Qm9x8172938475610293847561029384/2026-10-28-ocean-waves.mp4"
}
]
}
}The cost_micro_usd value shows the exact charge ($0.54 in this example). Because this was called under a project, the URL is permanent and expires_at is null.
Animating stills with image-to-video
You can also use an existing still image as the starting frame. This is useful for animating product photography, portraits, or artwork generated through image generation APIs.
Add image_url inside the input object:
{
"type": "video",
"model": "vibiz/video",
"input": {
"prompt": "Camera zooms out slowly as steam rises from the coffee cup",
"image_url": "https://thelastapi.dev/f/Qm9x8172938475610293847561029384/coffee-still.png",
"aspect_ratio": "16:9"
}
}From the CLI, pass the --from-image flag:
vibiz video "Steam rising from the cup, gentle camera drift" \
-from-image "https://thelastapi.dev/f/Qm9x8172938475610293847561029384/coffee-still.png" \
-o coffee.mp4Video specifications and parameters
vibiz/video standardizes video output across common formats so you do not have to manage resolution inconsistencies in your frontend video players.
| Parameter | Supported values | Details |
|---|---|---|
model | vibiz/video | Runs LTX-2.5 Fast (Lightricks) |
| Duration | 6 seconds | Fixed duration per render |
| Resolution | 720p | High-definition output |
| Audio | Included | Generated sound synchronized with the clip |
aspect_ratio | 16:9 or 9:16 | Horizontal or vertical; 1:1 is automatically returned as 16:9 |
image_url | Valid public HTTPS URL | Optional starting frame to animate |
| Pricing | $0.54 per video | Billed per successful job |
For applications supporting both desktop web feeds and mobile social interfaces, use 16:9 for widescreen players and 9:16 for mobile vertical stories. If you need unified access across still images, speech, and video under a single system, see our guide on one API for image, video and voice AI models.
Accessing specialized video models by id
Beyond the standard vibiz/video tier, you can call specialized models from other labs by passing their exact model id to POST /v1/ai/jobs. This gives you access to models like Kling, Seedance, Veo, Hailuo, or Wan without creating accounts with each research lab.
To search for available video models and check their live pricing, run:
vibiz media-models --category image-to-videoTo call a lab model directly, provide its id and pass its specific configuration inside the model_input object (up to 16 KB):
import { vibiz } from '@vibiz/app';
const job = await vibiz.ai.jobs.create({
type: 'video',
model: 'kling-video/v2.1/master/image-to-video',
input: {
prompt: 'A gentle breeze blowing through green grass',
image_url: 'https://thelastapi.dev/f/Qm9x8172938475610293847561029384/field.png',
model_input: {
duration: '10',
},
},
});Lab models called by id charge live rates against your account credits. To prevent accidental overspending, any individual call that would exceed $10 is rejected with a call_too_expensive error. For more details on pricing differences between tiers, consult our breakdown on how much AI video generation costs per clip via API.
Safeguarding production with budgets and spend limits
Video generation costs significantly more per call than text generation. Building video tools without strict limits can deplete an account balance quickly if users generate clips in tight loops.
thelastapi applies budget ceilings at multiple levels:
- Daily account spend ceiling: New accounts have a $5 daily limit across all AI calls and actions by default. An owner can raise this in the Billing console.
- Daily project budget: Each project defaults to a $1 daily AI budget. You can adjust this to fit expected production traffic:
vibiz ai budget 25.00 --project prj_123- Per-user tracking: Pass your internal user identifier in the
userparameter (orX-Vibiz-Userheader) with job requests. The console breaks down spend by end user, helping you identify heavy accounts.
When a job exceeds a budget, the API answers with an HTTP 429 error carrying details.reason: "project_budget" or details.reason: "daily_spend". Your code can inspect this error to inform users when their daily generation quota has been met.
Automated file storage
Handling video uploads manually requires setting up media buckets, generating signed upload URLs, and cleaning up failed uploads. thelastapi automates this entirely:
- Every video created under a project key is automatically ingested into the project's Files storage.
- The URL returned in the completed job payload is public and permanent, formatted as
https://thelastapi.dev/f/<key>/<filename>.mp4. - Clips generated using development keys are stored in the project's development Files environment, keeping test assets separate from production user data.
You can inspect, search, and download your generated clips using the CLI:
# Get the URL of the most recently generated video
vibiz files latest --type video
# Search for videos matching a specific prompt
vibiz files search "autumn leaves"If you call the API using an account key without specifying a project, the generated video link remains active for 24 hours. To make assets permanent, always link the generation call to a project.
To begin integrating video generation into your application, check the API parameters in the AI docs or review our guide on generating images and video from Claude Code.
Frequently asked questions
How long does it take for a video job to finish rendering?
Video generation via `POST /v1/ai/jobs` is asynchronous. Most six-second renders finish within a couple of minutes, during which your application polls `GET /v1/ai/jobs/:id` until the status changes from `pending` or `processing` to `completed`.
What happens if I request a 1:1 aspect ratio for a video?
The `vibiz/video` model accepts `16:9` and `9:16` aspect ratios. If your request submits a `1:1` aspect ratio, the gateway processes the clip and returns it formatted as `16:9`.
Are generated video links permanent or do they expire?
When you generate a video using a project key or pass a project header, the video is saved directly to your project's Files with a permanent link that never expires. Calls made with an account key outside of a project receive links that expire after 24 hours.