nsfwapis.com
// Guide / Video

NSFW AI video API: a developer guide

An NSFW AI video API is an HTTP endpoint that turns a source image, and on some models a text prompt, into a short video clip from a model that doesn’t refuse adult content. This directory lists 3 providers that offer one: Crebots, UncensoredLLM and NanoGPT. They price video in two different ways, per clip and per second, so this guide works out what a 5-second and a 10-second clip cost on each, then covers audio, model choice and the long async jobs video generation needs.

What is an NSFW AI video API?

An NSFW AI video API is a web endpoint you call from your own code to generate short video clips, from a provider that allows adult output and a model that doesn’t block it. You send a request with a source image, a motion prompt and some settings. You get back a job ID, and some time later a video file.

The clips are short. The listed options produce a few seconds of video per request, and Crebots’s longest published audio tier is 20 seconds. Longer scenes are built by chaining clips or by using the last frame of one clip as the source image for the next.

A typical video request carries these fields, though names differ between providers:

  • Model: the ID of the video model, for example an uncensored Wan or LTX build.
  • Source image: a URL or uploaded file that becomes the first frame.
  • Prompt: what should move and how (“slow camera push-in, hair moving in the wind”).
  • Duration and quality or resolution, which set the price.
  • Seed, so you can reproduce a result, and sometimes a callback URL.

If you haven’t built the image step yet, the NSFW image generation API guide covers it. Most video pipelines start there.

Image-to-video vs text-to-video: what NSFW video generation APIs support

Every NSFW video generation API in this directory supports image-to-video. That is the workflow to plan around: first make or pick a still image, then animate it. Text-to-video, where the model invents the whole scene from a prompt, is less common among the listed providers.

Image-to-video has practical advantages for adult apps. The source image fixes the character, the outfit and the framing, so the clip stays consistent with what the user already approved. It also gives you a checkpoint for moderation: you can check the still image before paying for video.

NSFW AI video API providers in the directory and what they support
ProviderImage-to-videoAudioPricing basisPayment
CrebotsYesYesImage to Video · Image to Audio + VideoCrypto only
UncensoredLLMYesNoVideo, per secondCard or crypto
NanoGPTYesNoPrices not listed yetCard or crypto

UncensoredLLM also lists a model called Wan 3.0 Video Uncensored alongside its image-to-video models. Its name doesn’t say what input it takes, so check the model page in your account before you design a text-to-video feature around it. NanoGPT offers image-to-video too, but the directory doesn’t list its video prices yet.

A two-step workflow covers most text-to-video needs anyway: generate an image from the prompt, then animate it. You pay for one image plus one clip, and you get a still you can show as a preview while the video renders.

Which models run behind uncensored AI video APIs?

The video models in this directory come from two open-weight families: Wan and LTX. Both publish their weights, which is why uncensored versions can exist at all: anyone can fine-tune open weights on data the original release left out. The guide to uncensored LLMs explains those methods.

  1. Wan. Wan is a family of open video foundation models from Alibaba’s Wan team. The Wan 2.1 technical report describes 1.3B and 14B parameter models that handle both text-to-video and image-to-video, and says the 1.3B model needs only 8.19 GB of VRAM. UncensoredLLM lists Wan 3.0 builds at $0.200 per second, its most expensive video tier.
  2. LTX. LTX-Video comes from Lightricks and was built for speed. Its paper reports generating 5 seconds of 768 by 512 video at 24 frames per second in about 2 seconds on an H100 GPU, with a model of about 2 billion parameters. UncensoredLLM lists LTX-2.3 Image-to-Video Uncensored at $0.080 per second, its cheapest video model.

On UncensoredLLM, the Wan builds cost 2.5 times as much per second as the LTX build. Whether the extra cost buys better motion for your content is something only your own tests can tell you. Neither name says how the uncensored build was trained, so run the same source images through both before you pick one.

Crebots doesn’t name the model behind its endpoints in the directory. It sells by quality tier instead, which is simpler to price but harder to compare on model quality alone.

Can an NSFW video API generate audio?

Yes, but only one listed option does. Crebots’s image-to-audio + video endpoint returns a clip with sound, in lengths from 5 to 20 seconds at low and mid quality. The other video models in the directory return silent clips.

If you need sound from a silent model, you have two options. You can add a separate audio track yourself (music, ambient sound, or speech from a text-to-speech API) and mux it with a tool like FFmpeg. Or you can use an audio + video endpoint and accept its quality tiers and lengths.

Audio generated together with the video can follow what happens on screen. A separate track gives you more control and lets you reuse the same sound across clips. For a first version, silent clips with optional background music are the simplest to build.

How much does an NSFW AI video API cost? Per clip vs per second

NSFW AI video APIs in this directory cost from $0.150 to $1.000 for a 5-second clip on published rates. The two pricing models behave very differently as clips get longer.

  • Per second. UncensoredLLM multiplies the listed rate by the clip length. A 10-second clip costs exactly twice a 5-second clip.
  • Per clip. Crebots charges a fixed price per clip by quality and, on the audio tier, by length. Its prices rise slowly with length.

Here is a worked example for a 5-second and a 10-second clip on each priced option:

Cost of one 5-second and one 10-second NSFW video clip at published rates, USD
Provider and modelPricing basis5-second clip10-second clip
UncensoredLLMLTX-2.3 Image-to-Video Uncensored$0.080 per second$0.400$0.800
UncensoredLLMWan 3.0 Image-to-Video Uncensored$0.200 per second$1.000$2.000
UncensoredLLMWan 3.0 Video Uncensored$0.200 per second$1.000$2.000
CrebotsImage to Audio + Video, lowPer clip, by length$0.150$0.165
CrebotsImage to Audio + Video, midPer clip, by length$0.165$0.180

On Crebots’s audio tier, a 20-second low-quality clip costs $0.195, only 30% more than a 5-second one. Per second, that drops from $0.030 to $0.010. If your users want longer clips, per-clip pricing gets much cheaper. On per-second models, the cost of long clips grows in a straight line.

Crebots’s plain image-to-video endpoint is also priced per clip: $0.150 at low quality and $0.225 at mid / high quality. The directory doesn’t list its clip length, so check the duration in the provider’s documentation before comparing it per second.

Worked example: a monthly video budget

Budget video by generations, not by clips users keep. Video has more ways to fail than a still image: a limb bends the wrong way, a face drifts, or the camera does something odd. Every retry is billed.

Take an app that delivers 2,000 kept 5-second clips a month, with 1.5 generations per kept clip. That is 3,000 billed generations:

Monthly cost of 3,000 five-second video generations at published rates, cheapest first
Provider and modelPer clipMonthly cost
CrebotsImage to Audio + Video, low$0.150$450.00
CrebotsImage to Audio + Video, mid$0.165$495.00
UncensoredLLMLTX-2.3 Image-to-Video Uncensored$0.400$1,200.00
UncensoredLLMWan 3.0 Image-to-Video Uncensored$1.000$3,000.00
UncensoredLLMWan 3.0 Video Uncensored$1.000$3,000.00

Add the cost of the source images if you generate them. At the cheapest image rates in the NSFW image API category, that is a small share of the total next to the video.

Three habits keep the bill in check:

  • Approve the still first. Let users confirm the source image before you spend money animating it.
  • Default to the short clip. On per-second pricing, a 10-second default doubles your cost. Offer longer clips as an option, not the default.
  • Cap free usage hard. One video costs as much as dozens of images, so free tiers that allow video need tight limits.

How async video jobs work

NSFW video APIs are asynchronous. Generation takes far longer than for an image, so the provider returns a job ID straight away and makes the clip available later. Crebots documents a task endpoint to poll and optional callbacks.

  1. Upload or link the source image. Some endpoints take a URL, others a file upload or base64 data. Large images may need resizing to the model’s supported resolution first.
  2. Submit the job. Store the job ID, the user, the prompt and the price you expect to pay.
  3. Wait with a long timeout. Poll every 5 to 10 seconds, or register a callback URL. Allow minutes, not seconds, before you call it a timeout.
  4. Copy the file. Download the finished clip to your own storage and serve it from there. Provider result URLs often expire.
  5. Handle failures. Treat “failed” as final, show the user a clear message, and decide whether failed jobs are retried or refunded in your own credits.

A minimal Python loop with placeholder paths, because they differ between providers:

import os, time, requests

BASE = os.environ["VIDEO_API_BASE"]
HEADERS = {"API-KEY": os.environ["VIDEO_API_KEY"]}

job = requests.post(f"{BASE}/SUBMIT_PATH", headers=HEADERS,
                    json={"image_url": "...", "prompt": "...", "duration": 5}).json()
deadline = time.time() + 600
while time.time() < deadline:
    status = requests.get(f"{BASE}/STATUS_PATH/{job['id']}", headers=HEADERS).json()
    if status["status"] in ("completed", "failed"):
        break
    time.sleep(8)

if status["status"] == "completed":
    clip = requests.get(status["video_url"], timeout=120)
    open(f"{job['id']}.mp4", "wb").write(clip.content)

Header names and response fields are examples. Crebots uses an API-KEY header, while OpenAI-compatible APIs usually take a bearer token. Check which one your provider’s video endpoint expects. In production, run this in a background worker, not inside a web request.

Consent, rights and moderation for video

Image-to-video raises the same risks as image-to-image, with higher stakes, because a moving clip of a real person is more convincing than a still.

  • Minors. Sexual content involving minors is illegal everywhere, including AI-generated video in many countries. Block it in prompts, source images and outputs. UncensoredLLM says it refuses any prompt involving a minor.
  • Real people. Animating a photo of a real, identifiable person into sexual content without their consent is illegal in many places and against every provider’s terms. Check uploaded images as well as prompts.
  • Rights to the source image. Only animate images you own or have permission to use. UncensoredLLM states that edits must use images you have the rights to edit.
  • Safety checkers. UncensoredLLM says its media models with a safety checker setting default to off. Filtering is your job.
  • Age checks and records. Adult platforms usually need age verification and may have record-keeping duties. Check the law where you and your users are, and each provider’s terms.

Check the source image before you submit a video job, sample frames from the finished clip, keep logs, and give users a way to report content. NSFW APIs hasn’t endpoint-tested any provider, so read each acceptable-use policy for your exact use case.

Checklist for choosing an NSFW AI video API

  1. Decide on audio. If you need sound in one pass, Crebots’s audio + video tier is the listed option. Otherwise plan to add audio yourself.
  2. Fix your default clip length. Then price it on both per-second and per-clip models with the table above.
  3. Test 20 to 30 of your own source images on two or three models. Note motion errors and how many attempts each needs per usable clip.
  4. Check payment. UncensoredLLM and NanoGPT take cards and crypto. Crebots takes crypto only.
  5. Build the job queue first. Polling or callbacks, timeouts, file storage and failure handling.
  6. Plan moderation before launch. Upload checks, output sampling and user reports.

Frequently asked questions

What is the cheapest NSFW AI video API?

For a 5-second clip on published rates, Crebots’s image-to-audio + video endpoint at low quality is the cheapest listed option at $0.150 per clip, and it includes sound. For longer clips, per-clip pricing on Crebots gets cheaper per second, so compare at the length you actually need.

Is there an NSFW video generation API with audio?

Yes. Crebots offers image-to-audio + video for clips of 5 to 20 seconds, from $0.150 per clip. The other video models in this directory return silent clips.

Is there an uncensored AI video API for text-to-video?

The NSFW video APIs in this directory are built around image-to-video. For text-to-video, generate an image from the prompt first and then animate it, or check UncensoredLLM’s model list for models that accept a prompt alone.

How much does a 10-second NSFW AI video cost?

On published rates, a 10-second clip costs $0.800 on LTX-2.3 Image-to-Video Uncensored at UncensoredLLM, $2.000 on its Wan 3.0 models, and $0.165 with audio on Crebots’s low-quality tier.

Why does an NSFW video API take so long to respond?

Video generation takes much longer than text or images, so the API queues the job and returns an ID right away. You poll a status endpoint or receive a callback, then download the finished clip.

Sources

Prices and offers come from each provider’s published information, checked . They can change, so confirm them on the provider’s site. Found something out of date? Tell us.