PixVerse NSFW: What V6 and V6 Max Refused, and What to Use Instead

PixVerse does not offer an unfiltered mode for NSFW generation. Its rules prohibit this category, and its API checks prompts, uploaded images and finished videos. Moving from the app to an API does not remove PixVerse's checks. We have not tested the app itself, so we cannot compare its refusal threshold with the API's.

In our tests on SpicyAPI, V6 refused the one Moderate text-to-video request we ran. V6 Max delivered Mild and Moderate text-to-video, but refused the Strong and Very strong requests on both text and image input. Effects has no intensity test results. If those limits stop your project, choose another model rather than search for a wording trick.

What PixVerse's rules mean for your project

PixVerse's Community Guidelines put this category on their prohibited list. Their enforcement options range from warnings to permanent bans. The website and app terms also allow access to be terminated for violations.

The API is not an exception. PixVerse's Platform Terms of Service apply restrictions to API use too. Section 3.3 prohibits attempts to evade its review mechanism; the terms allow API access or keys to be suspended or terminated. A clip completing in a test does not change those terms.

The checks also reach more than the prompt. PixVerse's error documentation distinguishes text, image, audio and video rejection. Its task-status documentation includes a moderation-failed state after generation. Replacing a text description with a source image therefore does not move the job outside the checks.

For a creator, the useful distinction is between a request the model completes and a service designed to run without this filter. V6 Max completed some requests further along our scale than V6, but neither is an unfiltered choice. Pick it for the tasks it handles within its rules.

What each PixVerse version actually delivered

These results come from tests on 25–26 October 2026 using fictional adults only. We use five intensity levels: Mild, Moderate, Strong, Very strong and Maximum; test prompts are not published. The intensity runs used 360p with sound off, and every returned result was visually checked.

“Delivered” means the clip followed the request. “Toned down” means it completed but did less than asked. “Refused” means no clip was delivered. These are small samples, not promised success rates.

V6: one refusal, several untested tasks

PixVerse V6 text-to-video refused the Moderate request, one run out of one. We did not intensity-test its Mild, Strong, Very strong or Maximum levels.

The other V6 tasks—image-to-video, first-and-last-frame animation, reference-to-video and extension—were functionally verified with page-safe examples. They were not intensity-tested. The family carries a Filtered label, but that label must not be mistaken for a completed test of every task.

This matters if you already have a source image: we cannot infer its outcome from the text refusal. We can say that image input remains subject to PixVerse's checks, not that we measured the same threshold on every V6 task.

V6 Max: text and image input have different limits

PixVerse V6 Max text-to-video went further than V6 on the Moderate text request. The corresponding image task did not.

The Moderate text result was a one-second clip. It establishes that this request completed, not that every longer request at that level will do so.

We also ran two different Maximum requests per task: one completed and one was refused. The completed request tested a different category, so it does not establish support beyond the Strong and Very strong requests that were refused. Do not read it as a path past those limits. Outcomes at the edge can vary between requests.

Nothing came back toned down in these V6 Max tests: each request either completed as written or was refused before delivery. The Mild image-to-video result went slightly further than requested, which is another reason to review a clip before using it. The V6 Max review covers the wider feature experience.

Effects: templates, not an unfiltered edition

PixVerse V6 Effects offers 50 themed templates, including sports celebrations, product presentations, dance and transformations. Of the 47 page-safe template examples we submitted in our functional test, 46 returned a finished clip.

That is not intensity coverage: none of the five levels was tested on Effects. It carries a Filtered label, and the template collection is intended for page-safe work. Choose it when you want an existing effect applied to a photo, not as a substitute for a model without a content filter.

A refusal, a failed input and a bill are different things

On SpicyAPI, a PixVerse content refusal is reported as content_rejected. For the V6 Max input refusals, the error identifies whether the prompt or image was flagged. Those requests were rejected in roughly one second; the V6 Moderate text refusal took about 20 seconds. A quick failure is therefore possible, but not guaranteed.

Failed or declined requests on these endpoints are not charged. That is SpicyAPI's billing behavior for these models; PixVerse's own API FAQ uses conditional refund language, so do not assume every service bills identically.

A malformed reference name is a different problem. V6 reference-to-video requires a space after an @name reference. In our functional test, punctuation immediately after the name caused invalid_request, not a content refusal. Fix that formatting error rather than treating it as evidence about the model's limits.

Is there a PixVerse prompt bypass?

There is no switch here that disables PixVerse's checks. Trying to evade the review mechanism conflicts with its API terms. We do not recommend rewriting a refused request to slip past the check, or using an image as a workaround.

The useful change is the model. A filter refusal calls for a service with a different content policy; a completed clip that does less than asked calls for a model that can deliver the result. Our guide to choosing the model before the words explains that distinction. Neither case is solved by assuming an API connection removes the model's own limits.

What to use when PixVerse refuses

These four alternatives are available on SpicyAPI. Prices below are USD for five seconds at 480p; they are starting configurations, not a comparison of identical visual quality.

Start with MiniMax H3 if you are replacing a V6 text or reference workflow. It covers the same broad set of input tasks and includes sound. Its tested source frames were generated images; real photographs were not tested.

Choose Wan 3.0 when you need sound, references or clips up to 30 seconds. Its stronger test coverage comes with a practical caveat: image and reference results sometimes added another person. Review the full clip, especially if the number of characters must stay fixed.

For a single still, Wan 2.2 Spicy is the lower-cost starting point among these four. It offers five- or eight-second clips and does not take text-only input. Seedance 2.5 Spicy is a more expensive image-based option with sound; its results are covered in the Seedance comparison.

All four offer extension and accept a last frame on image-to-video. For V6 Fusion-style references, use MiniMax H3 or Wan 3.0. There is no equivalent unfiltered Effects template family to recommend; switching from a template means describing the intended result yourself. The video model guide covers further choices.

Where PixVerse is still useful, and what it costs

For page-safe projects, V6 has useful controls: multi-shot generation, first and last frames, up to seven named reference elements and video extension. It generates one to 15 seconds, with optional sound and output up to 1080p. Extension charges only for newly generated seconds, although details can drift between the source and continuation.

V6 Max is the higher-quality edition, but it does not include V6's last-frame, reference, extension or multi-shot controls. Choose it for text or one starting image when those missing controls do not matter. Effects supplies a template's timing rather than a duration you choose.

Effects uses V6's per-second rates, multiplied by the template's actual delivered duration. Its initial hold covers up to 30 seconds, with the difference released after delivery. See the linked model pages for current prices before running a batch.

In our five-second 540p functional runs, V6's model time was 37–52 seconds for text, image, first-and-last-frame and reference generation; extension took longer. V6 Max took about a minute and a half. These timings exclude queuing and file transfer.

Features and inputs: Max trades controls for a simpler brief

V6 makes sense when a shot has requirements beyond “animate this picture.” A starting image determines the opening composition. An optional ending image tells the model where to arrive. In our observatory example, the person moved from the dome opening to the railing and finished in the supplied ending pose. That is a concrete reason to choose V6 even when you have budget for Max.

Named references solve a different problem. You can supply up to seven elements, each with a name and a role as a subject or background. The prompt then calls those elements with @name. Our functional example combined a person and a travel trailer, with the person walking toward the camera carrying a mug. This lets you specify who or what belongs in the scene without making a single finished starting frame first.

There is a small but consequential writing rule: put a space after each reference name. Our request using @trailer, failed because the name did not match; changing it to wording with @trailer and worked. If a reference request fails at this step, fix the spacing before changing the pictures or simplifying the entire scene.

Max accepts text or exactly one starting image. It has no ending-frame input, named-reference mode, extend task or multi-shot switch. That makes it a candidate for a self-contained shot whose composition is already settled. It cannot replace a V6 reference request simply by using the same prompt: the required input mode is absent.

On both families, image-to-video follows the starting picture's aspect ratio. Text-to-video offers eight shapes, including landscape, square, vertical and ultrawide. V6 also offers those shapes on reference-to-video. Prepare the image in the composition you want before animating it; a vertical instruction is not an aspect-ratio selector for an image task.

Limits that affect the edit

V6 extension returns the original and the continuation in one file. Our five-second source plus five new seconds became a single 10.1-second video, with billing for the new seconds only. At 540p without sound, five added seconds cost $0.1555 at the listed rate, rather than charging for the whole joined file.

The continuation preserved the coastal setting and motion, but the car changed color and shape. Use extension when continuing the action matters; be cautious when a recognizable product or vehicle must remain identical across the join. That observation comes from extending a V6 source, so it does not establish how every other source video behaves.

V6's multi-shot option cannot be combined with an ending frame. If landing on a specific composition matters, keep the ending image and use a single shot. If letting the model cut between moments matters more, use the multi-shot option without that frame. Max offers no equivalent switch.

For starting images, both accept JPG, JPEG, PNG or WebP up to 20 MB. Max also specifies a maximum of 10,000 pixels on either side. Neither family offers a fresh clip beyond 15 seconds in one request. Resolution tops out at 1080p; choosing Max does not add a higher tier.

Frequently asked questions

Does PixVerse allow NSFW?

Its rules prohibit this category, and checks apply on its API too. Some lower-intensity requests completed in our tests, but that does not make PixVerse an unfiltered service.

Is V6 Max less restrictive than V6?

On the Moderate text request we tested, yes: Max delivered and V6 refused. That observation does not extend to every task or level. Max image input was refused at Moderate.

Is V6 Max a replacement for V6?

It covers text-to-video and single-image animation. Keep V6 for ending frames, named references, extension or its multi-shot switch. A higher tier does not mean more input modes.

Does V6 Max have a higher maximum resolution?

No. Both offer 360p, 540p, 720p and 1080p. Max's positioning concerns picture quality, not an extra resolution option.

Can an uploaded image avoid a refusal?

No such workaround is established. V6 Max rejected source images as well as prompts, and its Strong and Very strong image requests were refused.

Can trying to evade the check affect API access?

PixVerse's API terms prohibit evasion and allow access to be suspended or terminated. Use a model whose policy fits the project instead.

For a text-first project, open MiniMax H3; for a low-cost still-to-video project, start with Wan 2.2 Spicy. Use fictional adults only, with no unauthorized real-person likenesses and no characters who could be mistaken for minors.