AI video API for developers · Updated November 2026
Nano Banana Video API
One REST API for three leading video models. Send a prompt or an image, pick video_model per job, get an MP4 URL back. No vendor juggling, no per-second bills.
The API at a glance
| Base URL | https://nanobananavideo.com/api/v1 |
| Authentication | Authorization: Bearer nb_… (key from your dashboard) |
| Endpoints | Text to video, image to video, video status |
| Models | MiniMax H3 (default), Seedance 2.5, FLUX 3 |
| Response | JSON with video_id, video_url (MP4) and thumbnail_url |
| Rate limit | 8 generations per 30 minutes |
| Price | €59.99 / month, flat |
Pick a model per request
Same request shape for every model: change one field to compare them on your own prompts and keep the winner.
| video_model | Model | Duration | Resolution | Best for |
|---|---|---|---|---|
minimax-h3 | MiniMax H3 Ultra Fast | 5–15 s | 480P, 768P | Default. Fast cinematic and commercial shots, native audio. |
seedance-2.5 | Seedance 2.5 Fast | 4–15 s | 480p, 720p | Social, UGC and e-commerce clips, auto audio. |
flux-3 | FLUX 3 Max | 5–20 s | 720p, 1080p | Highest fidelity, 1080p, longer takes. |
Aspect ratios: 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 (plus 2:1 on FLUX 3). Legacy ids veo3, seedance2 and flux3 keep working. Full matrix: model specs.
Endpoints
Your first request
curl https://nanobananavideo.com/api/v1/text-to-video \
-X POST --max-time 300 \
-H "Authorization: Bearer $NB_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "A matte sneaker rotating on wet slate, slow orbit, studio light",
"video_model": "minimax-h3",
"resolution": "768P",
"duration": 5,
"aspect_ratio": "16:9"
}'import os
import requests
resp = requests.post(
"https://nanobananavideo.com/api/v1/text-to-video",
headers={"Authorization": f"Bearer {os.environ['NB_API_KEY']}"},
json={"prompt": "Aerial shot of a lighthouse at dawn, waves crashing", "video_model": "flux-3", "resolution": "1080p", "duration": 8},
timeout=300, # renders take 1 to 3 minutes
)
resp.raise_for_status()
print(resp.json()["video_url"])const res = await fetch("https://nanobananavideo.com/api/v1/image-to-video", {
method: "POST",
headers: { Authorization: `Bearer ${process.env.NB_API_KEY}`, "Content-Type": "application/json" },
body: JSON.stringify({
image_url: "https://example.com/product.jpg",
prompt: "Slow push-in, soft studio light, the product turns slightly",
video_model: "seedance-2.5",
}),
signal: AbortSignal.timeout(300_000),
});
const data = await res.json();
if (!res.ok) throw new Error(`${res.status}: ${data.error}`);
console.log(data.video_id, data.video_url);Field reference and response shape: text-to-video docs and image-to-video docs.
How a generation works
- You POST a job. Prompt or image URL, model, duration, resolution and aspect ratio.
- The request waits for the render. Usually one to three minutes, so keep a client timeout of at least 300 seconds.
- You get the MP4. The response carries
video_url,thumbnail_urlandcredits_used. - Lost the connection? The render keeps going: read it with GET /video-status and its
video_id.
Rate limits and errors
The API allows 8 generations per rolling 30 minutes per account. Each generation response returns X-RateLimit-Remaining and X-RateLimit-Reset, so a queue can pace itself instead of hitting a wall. Errors share one shape, {"success": false, "error": "…"}, with the HTTP status telling you what to do:
- 401 missing or revoked key · 403 account not on the API plan
- 429 rate limit reached: wait for
Retry-Afterseconds - 500 the render failed: retry, or change the prompt or model
Full list: error reference · rate limits.
API or MCP?
Use the REST API when your own code triggers videos: an app, a batch of product clips, a pipeline. If you would rather ask for videos in plain language from Claude, ChatGPT or Cursor, connect the Nano Banana MCP server instead: same models, studio plan credits, no API plan needed.
What 300 API videos / month actually costs
The REST API is a flat €59.99 / month (API / Build plan). Generations are not billed per clip. Throughput is capped at 8 requests per 30 minutes per API key. MCP uses studio credits and does not require this plan.
Example: 100 product videos need 13 windows of 8 jobs, about 6.5 hours if you respect Retry-After. See batch generation and rate-limit docs.