Choose a video model
Pick the video model that makes the clip you need.
- Name the clip: its length, size, shape, sound and what it starts from.
- Read the video models and keep those that make it.
- Compare the ones left.
- Check what it charges for on the model's page.
1. Name the clip
A model makes a clip only in the lengths, sizes and shapes it lists. Settle these five before you look at a single model.
| The clip's | Sent as | Example |
|---|---|---|
| length, in seconds | duration | 10 |
| size | resolution | 1080p |
| shape | aspect_ratio | 9:16, for a phone story |
| sound | generate_audio | true |
| starting picture | frame_images | a first frame, a last frame, or both |
Every field of the request: Clips.
2. Read the video models
GET /v1/videos/models lists every video model on sale, with what each can make. Reading it is free. The sample keeps the models that make a 10-second clip at 1080p in 9:16 from a first frame, and prints each one's ID with the provider options it takes.
import json
import os
import urllib.request
API_KEY = os.environ["OPENDUNES_API_KEY"]
ADDRESS = "https://opendunes.com/v1/videos/models"
RESOLUTION = "1080p"
ASPECT_RATIO = "9:16"
FRAME = "first_frame"
DURATION = 10
request = urllib.request.Request(
ADDRESS,
headers={"Authorization": "Bearer " + API_KEY},
)
with urllib.request.urlopen(request) as response:
models = json.load(response)["data"]
def makes(model, field, value):
return value in (model.get(field) or [])
for model in models:
if (
makes(model, "supported_durations", DURATION)
and makes(model, "supported_resolutions", RESOLUTION)
and makes(model, "supported_aspect_ratios", ASPECT_RATIO)
and makes(model, "supported_frame_images", FRAME)
):
print(model["id"], model.get("allowed_passthrough_parameters"))The cURL tab filters with jq.
| Field | Holds |
|---|---|
supported_durations | the clip lengths it makes, in seconds |
supported_resolutions | the sizes it makes |
supported_aspect_ratios | the shapes it makes |
supported_frame_images | the frames it starts or ends on: first_frame, last_frame |
generate_audio | whether it makes the clip's sound |
seed | whether a seed makes a repeat likelier to match |
allowed_passthrough_parameters | the provider's own options it takes: Provider options |
A field is null where the model does not say.
No field says whether a model takes reference pictures. Its description does: Reference pictures lists the models that name them.
3. Compare the ones left
The two tables below were read from the video models list on 2026-09-29. Models come and go, so run step 2 before you rely on a row.
The clip's shape
| The clip needs | Models that make it |
|---|---|
| more than 15 seconds | alibaba/wan-3.0 and alibaba/wan-3.0-prime (up to 30) · bytedance/seedance-2.5 (up to 30) · black-forest-labs/flux-3-video (up to 20) · openai/sora-2-pro (up to 20) |
| 4K | google/veo-3.1 · google/veo-3.1-fast · bytedance/seedance-2.0 |
| 21:9, wide | bytedance/seedance-2.0, -2.0-fast, -2.0-mini, -2.5, -1-5-pro · alibaba/happyhorse-1.0, -1.1 · black-forest-labs/flux-3-video · minimax/hailuo-3, -3-max |
| 9:16, a phone story | every model that makes a clip from words, but minimax/hailuo-2.3, which makes 16:9 only |
| no sound of its own | minimax/hailuo-2.3 · minimax/hailuo-3-max · runway/gen-4.5; alibaba/happyhorse-1.0, -1.1 and both x-ai/grok-imagine-video models do not say |
What the clip starts from
| The clip starts from | Models that take it |
|---|---|
| words only | openai/sora-2-pro takes no frame |
| a first frame only | alibaba/wan-3.0, -3.0-prime, -2.6 · alibaba/happyhorse-1.0, -1.1 · minimax/hailuo-2.3 · runway/gen-4.5 · x-ai/grok-imagine-video, -1.5 |
| a first and a last frame | google/veo-3.1, -fast, -lite · kwaivgi/kling-v3.0-pro, -std, kwaivgi/kling-video-o1 · bytedance/seedance-1-5-pro, -2.0, -2.0-fast, -2.0-mini, -2.5 · alibaba/wan-2.7 · black-forest-labs/flux-3-video · minimax/hailuo-3, -3-max |
| an existing clip, or one photo to animate | black-forest-labs/flux-video-edit, black-forest-labs/flux-video-upscale, runway/aleph-2, heygen/avatar-iv: they edit, enlarge or animate, and list no clip length |
A model's own page in Models shows its description. Try two or three on the same prompt before you settle on one.
4. Check what it charges for
A clip is charged once, when its job reports completed. What counts differs by model, so read it on the model's page in Models, or in its pricing_skus on the list in step 2, before you settle.
| On these models, read 2026-09-29 | The charge counts |
|---|---|
| most | the seconds of video, at the rate of the size picked and, on some, of generate_audio |
kwaivgi/kling-v3.0-pro, -std · alibaba/wan-2.6 | the seconds, at one rate from words and another from a picture |
bytedance/seedance-1-5-pro, -2.0, -2.0-fast, -2.0-mini, -2.5 | the video tokens made |
x-ai/grok-imagine-video, -1.5 | the seconds, and each picture sent in |
minimax/hailuo-3 | the seconds, and each reference picture |
Sending a job and asking where it stands are free. A job that ends failed, cancelled or expired charges nothing. Where to read the charge: Video.
On a model charged by the second, the length drives the price. Make short drafts at a small size, then the final length and size.
Next: Text to video.
Last modified on