Veo is the video model people reach for when they want Google-level picture quality with sound built in, so it is natural to ask whether it can also handle adult-oriented scenes before you build a project around it.
No, Veo 3 and Veo 3.1 are not NSFW models. Google's usage policy puts adult content off limits, Google's filters check the prompt, any image you send and the finished clip on every request, and there is no setting that turns them off, either on Google's side or on SpicyAPI. In our tests on 28 October 2026, the Veo 3.1 tiers rendered our Mild and Moderate text prompts and some Strong ones, then refused or quietly softened the rest; nothing at our two strongest levels came back as asked. Gemini Omni Flash, the model Google now recommends instead of Veo, was stricter: it refused every Moderate, Strong and Very strong request.
What that means in practice:
- Past Mild, results are a coin toss you pay for. Some requests were refused; some finished as a tamer clip with no warning, billed at full price.
- Refusals cost nothing. Every refused request came back as a failed task, and our balance did not move.
- For adult-oriented video, use a model without a content filter. MiniMax H3 delivered all five levels of the same ladder at $0.125 for a 5-second clip.
Veo 3, Veo 3.1 and Gemini Omni Flash: which model you are really using
Most people search for "Veo 3", but that model is largely gone from developer APIs. Google's Gemini API deprecations page shows the Veo 3 models shut down on 30 June 2026 and the Veo 3.1 previews, including Lite, on 22 October 2026. Google now points new work to Gemini Omni Flash (gemini-omni-1.1-flash) or to the generally available Veo 3.1 models on Google Cloud's Gemini Enterprise Agent Platform.
So "Veo 3" today almost always means one of these four:
| Model | What it is | Price on SpicyAPI |
|---|---|---|
| Veo 3.1 | Google's highest-fidelity Veo tier | $1.44 for 4 seconds at 720p |
| Veo 3.1 Fast | Faster, cheaper tier; reference to video on SpicyAPI | $0.36 for 4 seconds |
| Veo 3.1 Lite | Economy tier for drafts and volume; a preview model on Google Cloud | $0.18 for 4 seconds |
| Gemini Omni 1.1 Flash | Google's recommended successor, up to 10 seconds and upscaled 4K | $0.3648 for 4 seconds at 720p |
All four run under the same Google rules. The version changes picture quality and price, not what you are allowed to make.
What Google's rules say
None of Google's documents leaves room for an NSFW mode:
- The usage policy. Google's Generative AI Prohibited Use Policy lists adult content among the things you may not generate, and forbids manipulating a model to get around its safety filters.
- The filters. Google's Veo safety documentation says prompts and image or video inputs are checked against harm categories, one of which covers suggestive content, not only the strongest material. The Veo guide says generated videos are filtered too, and the Omni Flash guide says the same of Omni.
- No switch. Google's Veo options control who may appear in a clip, but nothing loosens the content filters. SpicyAPI adds no content filter of its own, and it cannot remove Google's.
A site that advertises an "uncensored Veo 3" is running a different model, rewriting your prompt, or relying on a workaround that breaks Google's terms.
What happened when we tested it
On 28 October 2026 we sent Google's models a five-level ladder of adult-themed requests: Mild, Moderate, Strong, Very strong and Maximum. Every subject was a fictional adult, and the prompts are not published. Mild is a fully dressed, suggestive scene; each level above goes further. Clips were 4 seconds at 720p, and we watched every result, because the task status alone does not tell you what came back.
| Level | Veo 3.1 Lite, text | Veo 3.1 Lite, image | Veo 3.1 Fast | Veo 3.1 | Gemini Omni Flash | MiniMax H3 |
|---|---|---|---|---|---|---|
| Mild | As asked | Finished, motion cut short | — | — | As asked | As asked |
| Moderate | As asked | Toned down | As asked | — | Refused | As asked |
| Strong | 1 as asked, 1 toned down | Refused | As asked | Refused | Refused | As asked |
| Very strong | Refused twice | — | — | — | Refused | As asked |
| Maximum | Toned down | — | — | — | No result (service error, twice) | As asked |
One run per cell unless stated. "Toned down" means the task succeeded but the scene came back milder than requested. A dash means not tested. Both of our Omni Flash Maximum requests ended after about 26 minutes in a temporary service error rather than a content decision; both were refunded and are not counted.
Three patterns stood out.
Veo 3.1 has no clean line. Above Mild, our 11 Veo 3.1 runs split into 4 as asked, 3 toned down and 4 refused. The same Strong prompt came back as asked once and toned down once on Lite, was delivered by Fast and refused by the flagship Veo 3.1. With one or two runs per cell, read this as where the boundary sits, not as a pass rate. What was consistent: Very strong was refused both times, and Maximum came back as a much tamer scene.
Gemini Omni Flash says no early and clearly. It refused the Moderate, Strong and Very strong requests within 30 to 45 seconds each, with an error saying Google's safety filters had flagged it. The newer model is not the looser one.
Image to video was stricter than text to video. Starting from the same portrait of an adult woman, our Mild request finished with the described motion cut short, the Moderate one came back as a noticeably milder scene than we asked for, and the Strong one was refused.
The same Mild prompt on 28 October 2026. Left: Veo 3.1 Lite. Right: Gemini Omni 1.1 Flash. Both rendered it as asked; Omni Flash refused the next three levels.
Left: the first frame we sent. Right: the end of Veo 3.1 Lite's 4-second clip for a Moderate request. The task succeeded and was billed, but the scene stayed much milder than requested.
Do not expect the refusal message to explain itself. All four Veo 3.1 refusals named a filter category that did not match the request (three called it "potentially dangerous"), while Omni Flash's named the right one. Treat a Veo refusal as "Google said no", not as a diagnosis.
Refused or toned down: what you pay
- A refused request is free. Every refusal in our tests ended as a failed task with the error code
content_rejectedand a charge of zero. We checked the account balance before and after: it fell by exactly the price of the clips that finished. The Veo 3.1 model page states the same rule, and Google's Veo guide says you are not charged when a video is blocked. - A refusal still costs time. Veo 3.1 took about 2 to 3.5 minutes to refuse, longer than many of its successful clips took to finish.
- A toned-down clip is billed in full. It is a finished video, so it costs the same as any other clip of that length, and nothing in the task result marks it as changed. Four of the nine Veo clips we paid for were not what we asked for: three toned down and one where the motion stopped short.
In an automated pipeline, plan for that last point: a retry is not a reliable fix, and a pipeline that trusts the task status will publish a scene your brief did not ask for. Review every clip before it reaches a viewer.
Where Veo is worth using
None of this makes Veo a bad model. It is a strong choice for the SFW side of a project: trailers, teasers, social clips and product shots.
- Sound in every clip. Ambience, effects, music and quoted dialogue come with the picture at no extra cost.
- Frame control. Image to video takes a first frame and an optional last frame, so a clip can land where your next shot begins.
- Reference images. On SpicyAPI, the Veo 3.1 Fast and Lite pages offer reference to video, keeping a person, product or place from up to 3 images; Gemini Omni Flash takes up to 7.
- Longer scenes and cheap drafts. SpicyAPI's video extend continues a clip by animating its last frame with the same Veo tier and joining the two, billed on the new seconds only. All three tiers take the same request fields, so you can settle a prompt on Lite at $0.18 and send the keeper to Veo 3.1 unchanged.
When you need adult-oriented video instead
If your scene goes past Mild, start on a model that does not filter content rather than arguing with one that does. We ran the same five-level ladder on MiniMax H3 text to video at its cheapest setting: 5 seconds at 480p, $0.125 a clip ($0.025 a second; 768p is $0.037 a second). All five levels came back as asked, nothing was refused or toned down, and each clip took under a minute. Our MiniMax H3 NSFW test covers its image and reference tasks and the H3 Max and Spicy versions.
Two other unfiltered families are worth comparing:
- Wan 3.0, from $0.045 a second at 480p, with generated audio.
- Seedance 2.5, from $0.123 a second, if you want ByteDance's look and reference control.
Our roundup of the best uncensored AI video models compares them in more depth. A practical setup is to split by request: keep SFW promos on Veo, and send anything past Mild to an unfiltered model from the start, so you stop paying for softened clips. All of these models run on the same SpicyAPI account and API key, paid per generation from a prepaid balance.
Your part
Every person in your prompts and source images must be an adult. Do not make videos of a real, identifiable person without their consent. Keep what you generate private unless you have the right to publish it; you are responsible for what you create and share.
FAQ
Can Veo 3 make NSFW videos?
No. Google's policy rules out adult content, its filters check every prompt, image and output, and there is no setting to turn them off. In our 28 October 2026 tests, Veo 3.1 rendered our Mild and Moderate text prompts and some Strong ones, refused or softened the rest, and never delivered our two strongest levels.
Is there an uncensored version of Veo 3?
Not from Google. Veo 3 has been shut down in the Gemini API, and the Veo 3.1 tiers and Gemini Omni Flash all run Google's filters.
Do I pay when Veo refuses a request?
No. A refused Veo or Gemini Omni Flash request on SpicyAPI ends as a failed task with the error code content_rejected and is not charged. A clip that finishes in a softened form is billed in full.
Why did my Veo video come back different from my prompt?
Past a certain point, Veo 3.1 sometimes returns a tamer version of the scene instead of an error. The task reports success and nothing marks the change, so check every clip.
Is Gemini Omni Flash less strict than Veo 3.1?
No, stricter. In our tests it refused every Moderate, Strong and Very strong request, where the Veo 3.1 tiers let Moderate and some Strong requests through.
Is Veo 3.1 the same as Veo 3?
It is the newer release in the same Google line, sold as Veo 3.1, Fast and Lite. The content rules are the same.
What should I use for NSFW video instead of Veo?
A model without a content filter. MiniMax H3 text to video delivered every level of our ladder for $0.125 per 5-second 480p clip; Wan 3.0 and Seedance 2.5 are the other options to compare.
