Runway Veo negativePrompt: 1,000 characters, and Sume's options
Runway's API takes an optional negativePrompt of up to 1,000 characters on three Veo models. Sume's video request lists no such field, so use positive wording.

Runway's API accepts an optional negativePrompt of up to 1,000 characters on veo3, veo3.1 and veo3.1_fast, and Sume's video request does not list a negative-prompt field at all, so on Sume you describe what to avoid inside the prompt itself.
Runway added the field on July 10, according to its API changelog, which says that omitting it uses Runway's default. The Sume side below comes from the video generation docs and the Video Router docs. Both vendor pages were read on 2026-10-03.
What exactly did Runway add to its Veo endpoints?
One optional string. It describes unwanted output, it is capped at 1,000 characters, and it exists only on the three Veo model ids. Other Runway video models in the changelog, such as gen4.5 or seedance2_5, are not listed as taking it.
The pricing page lists Veo by audio. At Runway's $0.01 per credit, the figures below are what a single second costs before you add anything else.
| Model id | With audio | Without audio | Dollars per second without audio |
|---|---|---|---|
| veo3.1 | 40 credits | 20 credits | $0.20 |
| veo3.1_fast | 15 credits | 10 credits | $0.10 |
Does Sume's video request have a negative prompt field?
Not in the documented parameter table. POST /v1/videos lists model, prompt, duration, resolution, aspect_ratio, size, frame_images, input_references, generate_audio, seed, callback_url and provider. There is no negative-prompt row, and the docs do not say that one is silently honored.
Two neighbours matter. The music docs say a non-empty negative_prompt returns HTTP 400 for that model, so Sume does reject the field somewhere. And for video, provider.options is the passthrough route, but in v1 allowed_passthrough_parameters is empty for every model and non-empty options are rejected. You cannot smuggle a provider's negative prompt through it.
So the safe reading is: do not send a negative prompt field on a Sume video job. Ask GET /v1/videos/models for the model you pin and trust only what it reports.
How do you carry a Runway negative prompt over as wording?
Fold each unwanted item into the main prompt as a positive instruction about what the frame should contain instead. This is the same habit as in the MiniMax H3 prompt guide post: a clean sky is easier for a model to follow than a list of banned clouds.
- Runway negativePrompt "text, watermark, extra fingers" becomes "clean frame with no overlaid text, hands with five fingers".
- Runway negativePrompt "camera shake" becomes "locked-off tripod shot".
- Runway negativePrompt "blurry" becomes "sharp focus on the subject throughout".
What does this change about comparing the two?
If a Runway Veo test depends on a long negative prompt, a Sume run of a different model with only a prompt is not a fair match. Keep the positive prompt identical in both, record which one used a negative prompt, and compare only the clips from runs that share that setting.
Sume bills list price times 1.25 on every video model, and the Runway rates above are Runway's own. See Runway's credits against Sume dollars for the conversion, and read the live per-model price from the catalog rather than from this table. Sume's docs here do not name a Veo model id, so this post makes no claim about whether one is listed today.
Is a negative prompt worth the extra characters?
Runway's changelog does not say how strongly Veo weighs the field, and this post has no benchmark to offer, so treat it as one more input to test, not a guarantee. The useful experiment is cheap: run the same prompt twice on veo3.1_fast at 10 credits per second without audio, once with a short negativePrompt and once without, and compare the two clips side by side.
A five-second pair costs 100 credits, or $1.00, at Runway's $0.01 per credit. If the clips are indistinguishable, you have saved yourself a field to maintain. If the negative prompt removed a recurring artifact, copy its meaning into your positive wording and rerun on the model you ship with.
Keep the list short. A thousand characters is a ceiling, and a paragraph of banned items tends to read like a second prompt that contradicts the first. Three or four items that map to defects you have actually seen in your own outputs beat an exhaustive blacklist.
Sources
Related posts
More in Models
- Seedance 2.5 timestamp-level editing: what Sume exposes today
ByteDance describes timestamp-level editing and 30-second clips for Seedance 2.5. See which of those fields Sume's Video Router accepts for seedance-2.5.
- Seedance missing from the Sume Videos panel picker: what to do
The Videos panel picker lists Auto, Video 1.0, Kling 3.0, Wan 3.0, MiniMax and Grok. Seedance runs through the API by id, or through Auto.
- Shortest AI video clip via API: 2 seconds on Wan 3.0 only
Sume's video catalog by minimum duration: Wan 3.0 takes 2 seconds, Omni 3, Seedance and Kling 4, MiniMax H3 5. Plus how to trim below a floor.
- Shortest AI video clip by model: 2, 3, 4 or 5 second floors on Sume
Wan 3.0 starts at 2 seconds; Gemini Omni Flash at 3; Seedance and Kling at 4; MiniMax H3 at 5. The minimum duration for each Sume video model.
Written by Sume