Provider options
Send a control that only one video provider has, such as a negative prompt.
- Find the options a model takes.
- Put them under the provider's name.
- Send the job.
- Keep a plain request beside it.
1. Find the options a model takes
The common controls, duration, resolution, aspect_ratio, generate_audio and seed, need no provider option: Clips lists them. Anything else is a provider's own option, under the provider's own name.
A model's allowed_passthrough_parameters on the video models list names the options it takes. Their values are in the provider's own documentation. Read on 2026-09-29:
| Models | Provider | Options it takes |
|---|---|---|
google/veo-3.1, -fast, -lite | google-vertex | negativePrompt, enhancePrompt, personGeneration, conditioningScale, aspectRatio |
kwaivgi/kling-v3.0-pro, -std | atlas-cloud | negative_prompt, cfg_scale |
kwaivgi/kling-video-o1 | atlas-cloud | negative_prompt |
alibaba/wan-2.6 | atlas-cloud | negative_prompt, enable_prompt_expansion, shot_type, audio, size |
alibaba/wan-2.7 | atlas-cloud | negative_prompt, prompt_extend, audio, ratio, last_image, video, videos, images |
bytedance/seedance-1-5-pro, -2.0, -2.0-fast | seed | watermark, req_key |
openai/sora-2-pro | openai | quality, style |
minimax/hailuo-2.3 | minimax | prompt_optimizer, fast_pretreatment |
alibaba/wan-3.0, -3.0-prime | alibaba | none |
The sample in Choose a video model reads every model's list, options included.
2. Put them under the provider's name
provider.options holds one object per provider, under the provider's name from the table. Only the options of the provider that serves the job are sent to it. Google's options go one level deeper, under parameters, as Google's own API takes them.
{
"provider": {
"options": {
"google-vertex": {
"parameters": {
"negativePrompt": "blurry, low quality, text on screen"
}
}
}
}
}3. Send the job
The sample asks google/veo-3.1-lite for 4 seconds at 720p in 16:9, with a negative prompt: what the clip must not show.
import json
import os
import urllib.request
API_KEY = os.environ["OPENDUNES_API_KEY"]
ADDRESS = "https://opendunes.com/v1/videos"
MODEL = "google/veo-3.1-lite"
PROMPT = "A white jasmine flower opens on a tiled courtyard wall in Algiers"
NEGATIVE = "blurry, low quality, text on screen"
request = urllib.request.Request(
ADDRESS,
data=json.dumps(
{
"model": MODEL,
"prompt": PROMPT,
"duration": 4,
"resolution": "720p",
"aspect_ratio": "16:9",
"provider": {
"options": {
"google-vertex": {
"parameters": {"negativePrompt": NEGATIVE},
},
},
},
}
).encode(),
headers={
"Authorization": "Bearer " + API_KEY,
"Content-Type": "application/json",
},
)
with urllib.request.urlopen(request) as response:
job = json.load(response)
print(job["id"], job["status"])The answer is 202, with the job's id and status pending. Wait for it and fetch the clip as in Text to video.
An option the provider does not know is dropped, with no error. A misspelt name costs you the control, not the job: check the clip, and the name against the table in step 1.
4. Keep a plain request beside it
Options belong to one provider. If your app can move to another video model, keep the common request apart and add a provider's options only when that provider's model is picked. The same clip on bytedance/seedance-2.0-fast, with none:
{
"model": "bytedance/seedance-2.0-fast",
"prompt": "A white jasmine flower opens on a tiled courtyard wall in Algiers",
"duration": 4,
"resolution": "720p",
"aspect_ratio": "16:9"
}The clip is charged in DADAThe Algerian dinar, the one currency the docs show.More on Balance and charges → once, when the job reports completed, for what its model's page in Models lists. A job that ends failed, cancelled or expired charges nothing.
Last modified on