Video on this key is billed by the second. Pick a resolution, pick a length, and the clip costs the two multiplied. Below, every uncensored video model in the live catalog is read into a table and worked out for 5, 10, 15 second clips. If you want to try the model before writing any code, the Media page in the console and the video generator page run the same build from a prompt box. If you want the number in a field before anything renders, that is what the API gives you, and the rest of this post walks through it.
Uncensored AI video generator rates, by the second
This table comes from GET /v1/catalog/media each time the page builds, last on 2026-11-16. Each rate covers one second of finished output and moves with resolution only. The right-hand columns are that rate times the clip length; no figure here was typed in.
| Model | Tier | Rate per second | 5 seconds | 10 seconds | 15 seconds |
|---|---|---|---|---|---|
| LTX-2.3 Image-to-Video Uncensored | resolution 480p | $0.08 | $0.40 | $0.80 | $1.20 |
| resolution 720p | $0.16 | $0.80 | $1.60 | $2.40 | |
| resolution 1080p | $0.24 | $1.20 | $2.40 | $3.60 | |
| Wan 3.0 Image-to-Video Uncensored | resolution 480p | $0.20 | $1.00 | $2.00 | $3.00 |
| resolution 720p | $0.40 | $2.00 | $4.00 | $6.00 | |
| resolution 1080p | $0.80 | $4.00 | $8.00 | $12.00 | |
| Wan 3.0 Video Uncensored | resolution 480p | $0.20 | $1.00 | $2.00 | $3.00 |
| resolution 720p | $0.40 | $2.00 | $4.00 | $6.00 | |
| resolution 1080p | $0.80 | $4.00 | $8.00 | $12.00 |
LTX-2.3 Image-to-Video Uncensored accepts clips from 3 to 20 seconds; leave duration out and you get 5. You pay the seconds you requested at the rate of the resolution you chose, and that amount is locked in the response that accepts the job, before any rendering starts.
That puts the range at $0.08 per second for resolution 480p, up to $0.24 per second for resolution 1080p. Only the resolution moves the price. Audio, aspect ratio and seed leave it where it is.
Pull the catalog yourself
No key is needed to read it. Every model row lists its pricing tiers, and each tier has a when object stating which settings it covers, so your code can price a clip before sending it. The page you are reading is rebuilt from that endpoint on a five-minute cycle.
curl -s https://api.unrestrictable.ai/v1/catalog/media \ | jq '.data[] | select(.id == "ltx-2.3-i2v-uncensored") | {id, unit, pricing}'100 clips, priced at each resolution
Take one job, 100 clips of 10 seconds on LTX-2.3 Image-to-Video Uncensored, and run it through every tier above. Find the row for the resolution you want; the last column is the total.
| Tier | Rate | One clip (10 s) | All 100 |
|---|---|---|---|
| resolution 480p | $0.08 | $0.80 | $80.00 |
| resolution 720p | $0.16 | $1.60 | $160.00 |
| resolution 1080p | $0.24 | $2.40 | $240.00 |
You pay up front, from $10, by card, WeChat Pay or crypto, and no plan is required. Bigger top-ups add bonus credit. The rows below match the bonus table the gateway publishes; the last column divides the credit, bonus included, by the per-second rate for resolution 480p, rounded down.
| You pay | Bonus | Balance | Video seconds |
|---|---|---|---|
| $200.00 | +5% | $210.00 | 2,625 |
| $500.00 | +10% | $550.00 | 6,875 |
| $1,000.00 | +15% | $1,150.00 | 14,375 |
Chat models on the same key bill per token; those rates are on the pricing page. Clips and chat requests share a single balance.
Submit, poll, download: the whole request
Every media model here, image or video, takes POST /v1/generations. The body names the model and carries an input object: the prompt plus the settings. You get a 202 straight back with a generation id and the charge, then poll that id until status reads succeeded or failed. Base URL and key are what the docs page hands every client.
curl https://api.unrestrictable.ai/v1/generations \ -H "Authorization: Bearer sk-unr-live-…" \ -H "Content-Type: application/json" \ -d '{ "model": "ltx-2.3-i2v-uncensored", "input": { "prompt": "a lighthouse at dusk, slow push-in, waves on the rocks", "resolution": "480p", "duration": 5 } }'This is the 202, returned before rendering begins. charge.units is the length you requested; charge.amount_usd is that length times the rate.
{ "id": "gen_7b2e41…", "object": "generation", "model": "ltx-2.3-i2v-uncensored", "kind": "video", "status": "queued", "progress": 0, "outputs": [], "error": null, "charge": { "amount_usd": 0.4, "units": 5, "unit": "second", "refunded": false }}Expect one to a few minutes. Ask again about every ten seconds; the status passes through queued and running on its way.
curl https://api.unrestrictable.ai/v1/generations/gen_7b2e41… \ -H "Authorization: Bearer sk-unr-live-…" { "id": "gen_7b2e41…", "status": "succeeded", "progress": 100, "outputs": [ { "url": "https://api.unrestrictable.ai/v1/files/2026-11-16/gen_7b2e41…/0.mp4", "content_type": "video/mp4", "bytes": 4810213 } ], "charge": { "amount_usd": 0.4, "units": 5, "unit": "second", "refunded": false }}The file is at outputs[0].url, stored on this domain for 30 days, and the same key downloads it. When a job fails, status is failed, error.message says why, charge.amount_usd reads 0 and charge.refunded is true. You are never billed for a clip that did not render.
Here is the loop in Python with requests.
import time, requests BASE = "https://api.unrestrictable.ai/v1"HEADERS = {"Authorization": "Bearer sk-unr-live-…"} gen = requests.post( f"{BASE}/generations", headers=HEADERS, json={"model": "ltx-2.3-i2v-uncensored", "input": {"prompt":"a lighthouse at dusk, slow push-in, waves on the rocks","resolution":"480p","duration":5}}, timeout=30,).json()print("charged", gen["charge"]["amount_usd"]) # locked in now, before rendering starts while gen["status"] not in ("succeeded", "failed"): time.sleep(10) gen = requests.get(f"{BASE}/generations/{gen['id']}", headers=HEADERS, timeout=30).json() if gen["status"] == "succeeded": url = gen["outputs"][0]["url"] # stays on https://api.unrestrictable.ai for 30 days with open("clip.mp4", "wb") as f: f.write(requests.get(url, headers=HEADERS, timeout=120).content)else: print(gen["error"]["message"], "refunded:", gen["charge"]["refunded"])Want to confirm the billed length against the file? ffprobe prints the duration; it should equal charge.units.
ffprobe -v error -show_entries format=duration -of default=nw=1:nk=1 clip.mp4# the clip length in seconds; it should equal charge.unitsLTX-2.3 Image-to-Video Uncensored: the full settings list
Read from the catalog row, so a model update changes this table too. This build has no stock prompt filter in front of it, and the enable_safety_checker flag stays false unless you set it.
| Name | Type | Accepts | Default |
|---|---|---|---|
prompt (required) | string | string | |
image_urls (required) | urls | urls | |
resolution | string | 480p, 720p, 1080p | 480p |
duration | integer | 3 to 20 | 5 |
seed | integer | 0 to 2147483647 |
LTX-2.3 Image-to-Video Uncensored accepts image_urls alongside the prompt, so you can start the clip from a reference frame. The page for the model itself, ltx-2.3-i2v-uncensored, shows the same tiers with a one-line starter request.
Every uncensored media model here is held to two rules, checked before a job is accepted: no minors depicted, and no likeness of a real person who has not consented. Break the first and the request comes back 400 content_not_allowed with no charge. The full wording is on the acceptable use page.
What the web apps do better
Search Google for "uncensored ai video generator" (we did, on 2026-10-19) and the page is prompt-box web apps: Senzia, VideoAny, MagicLight, Crano and ZenCreator, with Reddit threads and roundup lists between them. Their strength is the first five minutes. Type a prompt, watch it render, save the file. The listings lean on exactly that: 60 fps at 1080p and credits with no end date from ZenCreator, "MP4 export" and "free to start" from Senzia. Nothing to install, no key, no polling.
Venice offers a single API covering text, image, video and audio, plus a video generator inside its app. Its pricing page, when we read it on 2026-10-19, quoted video per job through a quote API rather than at a flat per-second rate. Check that page for the figure; none is copied here.
The third path those results push is running a model yourself. The Reddit reply was to go offline, and one YouTube pick fit in 8 GB of VRAM. Own the GPU and have the hours, and each second costs nothing beyond the hardware.
What the uncensored video API does better
- You see the price first. $0.08 a second for resolution 480p, times your requested length, printed in the 202 before a frame exists. A credit pack or a daily quota never says what one clip came to.
- Billed by the second. A 3-second trial is billed as 3 seconds. Per-clip pricing charges the trial and the full render alike.
- Chat and media on one key. The
sk-unr-key you use against the models page also submits video, and both land on a single statement drawn from a single balance. - A prompt box as well. The Media page in the console drives the same model under one of your keys and keeps a list of every clip with its status and cost, and the video generator page is the no-code way in. The quick test needs no code here either.
- No charge for failures. A failed generation is refunded in full the moment it fails, and the record shows it.
What this page can compare fairly is price and turnaround. It does not grade output. The real test is one prompt, run here and on another service, with the two files side by side.
Image and video models on the same key
The media page lists the video build next to the uncensored image models; all bill per image or per second, and all take the same endpoint. The image side gets its own numbers in the post on the uncensored image generator API.
| Per | Endpoints | |||
|---|---|---|---|---|
| LTX-2.3 Image-to-Video Uncensored | Video | $0.08 | second | |
| Wan 3.0 Image-to-Video Uncensored | Video | $0.20 | second | |
| Wan 3.0 Video Uncensored | Video | $0.20 | second | |
| Anime Image Uncensored | Images | $0.03 | image | |
| Chroma Uncensored | Images | $0.08 | image | |
| CyberRealistic XL Uncensored | Images | $0.0136 | image | |
| Qwen Edit Uncensored | Images | $0.08 | image | |
| Qwen Image 2.1 Uncensored | Images | $0.0375 | image | |
| Qwen Image 3 Uncensored | Images | $0.08 | image | |
| Seedream 5.0 Lite Edit Uncensored | Images | $0.0313 | image | |
| Vision Image Uncensored | Images | $0.04 | image | |
| Vision Pro Image Uncensored | Images | $0.09 | image | |
| WAI Illustrious SDXL Uncensored | Images | $0.0136 | image |
Questions
The consumer sites on page one of that search (2026-10-19) offer free starts and credit packs, and open-weight video models cost nothing to run once you own the GPU. This API charges: you prepay from $10, and every clip's price, $0.08 a second for resolution 480p, is set before rendering. See the image and video page.
Venice has a single API for text, image, video and audio; on 2026-10-19 its pricing page quoted video per job through a quote API, so read it for today's number. Video here is billed per second at fixed resolution tiers, $0.08 to $0.24 per second, pulled live from the catalog.
The sites ranking for that phrase are consumer apps built around a prompt box. Here you get the API plus a prompt box in the console: the Media page drives the uncensored video build under one of your keys, listing each clip and what it cost. Both routes follow two rules: no minors in any clip, and no real person shown without consent. See the full rules.
$2.40 as of this render: 10 seconds times $0.24 a second, the current resolution 1080p rate. The figure is locked in the 202 that accepts the job, and a failed clip is refunded in full.
Yes. LTX-2.3 Image-to-Video Uncensored accepts image_urls beside the prompt, so a reference frame can open the clip.
Between one and a few minutes. The submit call returns an id and the charge immediately; poll GET /v1/generations/{id} about every ten seconds until status reads succeeded or failed. The finished MP4 is kept here 30 days. See the API docs.
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.