The uncensored image API and the video route, for report graphics

Two routes for the image and video models: the OpenAI image call and a generations endpoint you poll. Requests, price-setting fields, a poll loop, refunds.

Steps checked October 26, 2026 · prices as of November 17, 2026

Report covers, diagrams for a client deck, a short clip for an awareness briefing: the 13 image and video models in this catalog make them on the same key as your chat and agent requests. Two routes serve them: OpenAI’s image API for stills, and a single generations endpoint with polling for video and anything slow. Below are both requests, the settings that decide the price, a poll loop in Python and the refund rule, with each price read live from the catalog.

The two routes and when each one applies

POST /v1/images/generations
OpenAI’s image request
prompt as a top-level field, everything else under input; blocks for up to 100 seconds and returns OpenAI’s response shape
POST /v1/generations
{ model, input }
all settings live under input; returns 202 and a generation object to poll
GET /v1/generations/{id}
the generation
its status, each file with URL, type and size, and the charge in its units

The first route is there so client.images.generate from the OpenAI SDK needs no changes. Video, image-to-video, and any job likely to outlast the image route’s wait belong on the generations route. One key from the console covers both, the same key your chat and agent requests use.

An image through the OpenAI call

The shortest request against Anime Image Uncensored, here making a cover graphic for a report. response_format takes url or b64_json. If the job is still running at 100 seconds the route answers 504 generation_pending and hands back the generation id, which you fetch from the generations route.

curl
1curl https://api.pentestllm.ai/v1/images/generations \
2 -H "Authorization: Bearer sk-ptl-live-…" \
3 -H "Content-Type: application/json" \
4 -d '{"model": "anime-image-uncensored", "prompt": "cover illustration for a penetration test report, abstract network map, flat vector, dark blue", "response_format": "url"}'
python
1from openai import OpenAI
2
3client = OpenAI(base_url="https://api.pentestllm.ai/v1", api_key="sk-ptl-live-…")
4result = client.images.generate(
5 model="anime-image-uncensored",
6 prompt="cover illustration for a penetration test report, abstract network map, flat vector, dark blue",
7 response_format="url",
8)
9print(result.data[0].url)

Anything OpenAI’s request shape lacks a field for, such as the safety checker, the sampler or a seed, goes under input. The defaults for each setting are on the model’s page.

A clip through generations and a poll

Post the job to Wan 3.0 Video Uncensored, then fetch the generation until its status is succeeded or failed. Billing is per second, and duration is accepted from 2 to 30 seconds. The example asks for a title card for an awareness briefing.

curl
1curl https://api.pentestllm.ai/v1/generations \
2 -H "Authorization: Bearer sk-ptl-live-…" \
3 -H "Content-Type: application/json" \
4 -d '{"model": "wan-3.0-video-uncensored", "input": {"prompt": "title card for a security awareness briefing, padlock icon, slow zoom", "duration": 5}}'
curl
1curl https://api.pentestllm.ai/v1/generations/gen_<id> \
2 -H "Authorization: Bearer sk-ptl-live-…"

To animate a still, for instance a diagram from the report, LTX-2.3 Image-to-Video Uncensored wants the first frame as a public URL under image_urls:

curl
1curl https://api.pentestllm.ai/v1/generations \
2 -H "Authorization: Bearer sk-ptl-live-…" \
3 -H "Content-Type: application/json" \
4 -d '{"model": "ltx-2.3-i2v-uncensored", "input": {"prompt": "the network map lights up node by node", "image_urls": ["https://example.com/frame.png"]}}'

The poll loop in Python

A five-second sleep between fetches is enough. A failed generation comes back with status failed, the reason, and its charge marked refunded.

poll.py
1import time, requests
2
3H = {"Authorization": "Bearer sk-ptl-live-…"}
4gen = requests.post("https://api.pentestllm.ai/v1/generations", headers=H, json={
5 "model": "wan-3.0-video-uncensored",
6 "input": {"prompt": "title card for a security awareness briefing, padlock icon, slow zoom"},
7}).json()
8
9while gen["status"] not in ("succeeded", "failed"):
10 time.sleep(5)
11 gen = requests.get(f"https://api.pentestllm.ai/v1/generations/{gen['id']}", headers=H).json()
12
13print(gen["status"], gen.get("charge"), [f["url"] for f in gen.get("files", [])])

Image prices

The entry price of each image model and the tier it belongs to, read from the media catalog when this page rendered. Every tier is on the model page.

ModelUnitFromLowest tier
Anime Image Uncensoredper image$0.03resolution 1K
Chroma Uncensoredper image$0.08any settings
CyberRealistic XL Uncensoredper image$0.0136size 512x512
Qwen Edit Uncensoredper image$0.08any settings
Qwen Image 2.1 Uncensoredper image$0.0375resolution 1K
Qwen Image 3 Uncensoredper image$0.08any settings
Seedream 5.0 Lite Edit Uncensoredper image$0.0313any settings
Vision Image Uncensoredper image$0.04any settings
Vision Pro Image Uncensoredper image$0.09resolution 2K
WAI Illustrious SDXL Uncensoredper image$0.0136size 512x512

Video prices

A clip is charged per second of output at the tier its resolution and settings select, so five seconds cost five times that tier’s per-second rate.

ModelUnitFromLowest tier
LTX-2.3 Image-to-Video Uncensoredper second$0.08resolution 480p
Wan 3.0 Image-to-Video Uncensoredper second$0.20resolution 480p
Wan 3.0 Video Uncensoredper second$0.20resolution 480p

Which settings move the price

  • On most models size, aspect_ratio and resolution decide the tier; the price table on the model page says which values change it.
  • n asks for more than one image or clip in a single request, and each is charged at the unit rate.
  • For video, duration is the count of billed seconds.
  • enable_safety_checker is off by default on 7 of the 13 image and video models listed; pass true to switch it on for one request.

Charges, refunds and retention

  • A failed generation is refunded in full: the charge shows refunded and the balance is restored.
  • Output files are served from the API’s /v1/files path for 30 days. Pull anything you need for the report into your own storage.
  • A prompt that describes a minor is rejected before pricing and before any model sees it, with HTTP 400 content_not_allowed. The complete list is on the acceptable use page.
  • The generation object carries the charge for a non-streamed answer, and the usage page in the console itemizes each request with its units.

Questions

Yes. Set base_url to https://api.pentestllm.ai/v1 and pass a model id from the media page. Any setting OpenAI’s shape lacks goes under input.

POST /v1/generations with the model and an input object, then GET the generation until it says succeeded. Both calls and a Python loop are above.

No. The generation reads failed, the charge is marked refunded, and the amount returns to the balance.

Thirty days, as links under the API’s /v1/files path. Copy what you want to keep before then.

The media page lists every image and video model with its entry price, 13 of them at render, and each model page has the tiers. See the media page.

Every model above is on the uncensored model list with its live price per million tokens. Per-token pricing sets the whole catalog side by side, and the family directory groups the builds by base model.

Get a keyPrepaid from $10, no subscription.

New models and price changes, once a month.

New uncensored builds with their prices, price moves, and what the daily model check found.

Help