AI video negative prompt: how to keep things out of a clip

A negative prompt is a separate list of things a generated image or clip should not contain, on tools that offer such a field. Sume's video API has none: POST /v1/videos has no negative_prompt field, and today a request that sends one is refused with a 400 before any job starts. To keep things out of an AI video on Sume, describe the shot you want in the prompt, then pin what must not change with a first frame, reference images, or the audio switch.

The request fields come from Sume's Video generation docs and the request schema in the Sume API reference, read on 2026-09-27. Anything described as current behavior is read from Sume's API code.

Does Sume's video API accept a negative prompt?

No. None of the request parameters the docs list for POST /v1/videos is a negative prompt, and the request schema does not allow fields it does not define. The description goes in prompt, which the docs call a text description of the video to generate.

In the current code, an undefined field fails validation as 400 invalid_request with the message “Invalid request body.”, and details.errors names the problem. Validation runs before the job is submitted, so the refusal creates no job. Video generation API 400 errors covers the other refusals on this route.

How do I keep unwanted things out of an AI video?

Sume's docs do not say how video models read “no X” wording in a prompt. They do document request fields, and each one fixes part of the clip:

{
  "model": "seedance-2",
  "prompt": "One woman in a gray coat in front of a plain white wall turns to camera. Static medium shot, soft window light from the left.",
  "frame_images": [
    {
      "type": "image_url",
      "image_url": { "url": "https://example.com/first-frame.png" },
      "frame_type": "first_frame"
    }
  ],
  "generate_audio": false
}

What should the prompt say instead?

Say what should be in the frame. Sume's best-practice advice for video prompts is to be specific and descriptive, with details about motion, camera angles, lighting, and scene composition. Where an image tool's negative prompt might read “text, crowd, blur”, describe the frame you want instead: one person, a plain white wall, a static camera, soft light.

Sume does tell callers to put exclusions in the positive prompt, but for music. Its Music Router docs say a non-empty negative_prompt is unsupported, and the example prompt ends “Instrumental, no vocals.” The video docs have no such rule, so treat “no X” wording in a video prompt as something to test on your model, not a guarantee.

How do I test whether “no X” wording works?

Change only the wording and keep everything else fixed between runs: the same model, duration, resolution, and aspect_ratio, and the same first_frame if you use one, so every take opens on the same image. The docs make no promise about what appears after that frame, so review each clip before you use it.

Each run is still a new take, not an edit of the last one: no v1 video model accepts seed, so there is no documented way to replay a clip with one word changed. What is an AI video seed? covers what to hold fixed instead.