Tools
Wyzie Censored#
Wyzie Censored finds the words you choose in a video or audio file and bleeps them, mutes them or covers them with a soft sound. Speech recognition gives every word its exact time, so only that word goes quiet and the rest of the line plays as normal. You get a new file back with the picture untouched and the audio cleaned.
It is made for families watching together, classrooms, streamers who need clean clips, and anyone who wants a film without the swearing. Players that already have captions or subtitles can skip the file and use Censored Live, which tells the player when to go quiet while the video plays.
On the Web#
To try it without code, open sub.wyzie.io/censored, upload a video or paste a link, and watch the clean result right there. It shows the price before anything is charged and uses your account settings.
Account Settings#
Turn Censored on and choose your words at dash.wyzie.io, under your account and then Censored. Until it is on, Censored requests from that account's keys get a 403 with code: PRODUCT_OFF. Keys that do not belong to an account use the defaults below. Every setting can also be changed for a single request with the parameters further down.
{
"code": "PRODUCT_OFF",
"message": "Wyzie Censored is off for this account",
"details": "An owner or admin can turn it on in dash.wyzie.io (your account, then Censored). Nothing was charged.",
"product": "censor",
"enable": "https://dash.wyzie.io"
}If the settings were changed in another tab or session since you opened them, a save is refused with 409 (prefs_changed) instead of overwriting that change: reload the settings and make your change again.
Strictness Levels#
A level picks the word lists for you, from the lightest to the strictest. Choose one in the console under Strictness, or send level with a single request. A new account starts at Moderate.
| Level | Name | What Is Censored |
|---|---|---|
lenient | Lenient | Only the worst swear words and slurs (profanity_severe, slurs). |
moderate | Moderate | Swearing and slurs. Mild words such as damn and hell stay (profanity_strong, profanity_severe, slurs). |
strict | Strict | Moderate, plus damn, hell, crap and bloody, sexual words, and taking God's name in vain (adds profanity_mild, sexual, blasphemy). |
super_strict | Super Strict | Strict, plus gosh, darn and heck, and insults such as stupid, idiot and shut up (adds minced, insults). |
kids | Kids | Everything Super Strict censors: the same lists under a name for families. |
teen | Teen | Swearing, slurs, sexual words and cursing with God's name (goddamn). Damn, hell, crap and oh my god stay (profanity_strong, profanity_severe, slurs, sexual, blasphemy_curses). |
family | Family | Teen, plus crude mild words such as crap, piss and bloody. Damn, hell and oh my god stay (adds mild_crude). |
Kids, Teen and Family are presets named for who is watching. Kids has the same lists as Super Strict, so GET /censor/settings reports its level as super_strict.
If you tick lists by hand and they match no level, the console shows the level as Custom. Custom is only a label. It is never sent as a level value.
Word Lists#
Pick any of these lists, or let a level pick them. A new account starts with profanity_strong, profanity_severe and slurs, which is Moderate.
| List | Meaning |
|---|---|
profanity_mild | Mild swearing, such as damn, hell and crap. |
profanity_strong | Strong swearing and insults. |
profanity_severe | The harshest swear words. |
slurs | Slurs aimed at groups of people. |
blasphemy | Holy names used as swear words, such as oh my god. |
sexual | Sexual words and phrases. |
minced | Mild stand-ins for swear words, such as gosh, darn, heck and holy cow. |
insults | Mild insults and rude words, such as stupid, idiot, shut up and fart. |
Two lists also come in halves, for when you want to keep some of their words: turn on a half instead of the whole list. A word a half finds still reports the whole list as its category.
| List | Meaning |
|---|---|
mild_damn_hell | Half of profanity_mild: damn, dammit and hell. |
mild_crude | Half of profanity_mild: crap, piss, bloody, freaking and the like. |
blasphemy_exclamations | Half of blasphemy: oh my god, Jesus, good lord, geez and the like. |
blasphemy_curses | Half of blasphemy: goddamn and god damn it. |
Speech recognition sometimes hears a near-word as a swear word, such as "pitch meeting" as "bitch meeting". A word from the lists is left alone when the word right before or after it gives the real word away. Your own Blocked Words are always censored.
Lists in Other Languages#
23 languages have their own word lists: es fr de pt it nl pl ru uk tr ar hi ja ko zh id vi sv da no fi cs el. pt-BR uses the pt lists and zh-TW the zh lists. Other languages, such as en, he and th, use the English lists. GET /censor/settings lists the codes under available.languages.
Coverage keeps growing. The English lists recently gained about 270 words, and the lists in the other 23 languages are now 1.5 to 3 times their earlier size. The categories, what each one covers and the lists a level picks are unchanged.
A job picks the lists from its language option or, without it, from the language it hears. For POST /censor/mask, add language to the body (es, pt-BR); a bad code is a 400 language is a language code (es, pt-BR). The language lists cost nothing extra.
Blocked Words and Never Censor#
Add words under Blocked Words (up to 500). They are always censored, whatever the level. Add words that must never be hidden under Never Censor (up to 200). A Never Censor word wins over every list. Each entry is 1 to 64 characters, and case does not matter.
A plain blocked word also catches its English forms. It never catches a different word that starts or ends with the same letters: rape does not catch rapeseed or grape.
In English, a blocked word also catches the forms people really say and write: endings, compounds and common misspellings. For example, fart also catches farting, fartface and phart. Wyzie asks an AI model for a word's forms once and keeps the answer for every account, so it is free (up to 100 new words a day per account). The first job with a brand-new word may run with the rule forms only and the next one gets the full set; saving the word under Blocked Words in the console looks it up in the background. Everyday words that only share letters are never added, and Never Censor still wins. Other languages do not get these AI forms yet.
| Entry | Meaning |
|---|---|
rape | That word and its forms: rape also catches raped, raping, rapes and rapist. |
=ass | Only that exact word, no other forms: =ass blocks ass, not asses. |
damn* | Every word that starts with it: damn* also catches damned. |
*damn* | That text anywhere inside a word. |
shut up | A phrase of up to 5 words, matched exactly, word for word. |
Never Censor entries and the preset lists match exactly as written, with no extra forms.
Named Profiles#
Save up to 10 named profiles in the Censored settings on dash.wyzie.io, such as one for the kids and one for movie night. A name is 1 to 40 characters. Each profile has its own level (lenient, moderate, strict, super_strict, kids, teen, family or custom), lists, blocked words and Never Censor words.
Send profile=<name> to POST /censor, POST /censor/url or POST /censor/mask, in any case. The profile's lists replace the account's, and its blocked words and Never Censor words are added to the account's. An unknown name is a 400 that lists your profiles, and nothing is charged:
No Censored profile "movie night". This account has: Kids, Teen room.GET /censor/settings lists them under profiles, each with name, level, lists and how many words and allow entries it has. It never returns the words.
How It Sounds#
Choose what a hidden word sounds like: bleep plays a 1 kHz tone, mute leaves silence, and sound plays a gentler replacement sound. The sounds are chime, soft, static and whoosh. They are made on the fly, so every player hears the same thing.
Padding (pad_ms, 0 to 300, default 60) adds that many milliseconds before and after each word, so no word slips out at the edges. Two hidden stretches less than 120 ms apart are joined into one.
Loud Music Mode#
Loud music mode (music=1, off by default) is for films where music, effects or a second voice drown out the speech. Wyzie also listens to the voices with the music taken out, and to the quieter of two people talking at once, and censors the words it finds there too. It costs 1,315 more requests per started minute (1,925 in all), charged when the job starts. Of that, 675 is for the voices without the music: all 675 come back when the gaps between words turn out quiet (music: not_needed) or that pass fails (music: failed); 75 come back when the voices were already separated in the last 24 hours, and 600 when they were already recognised. The other 640 is for two voices at once (below). The job result says music: used, not_needed or failed. Turn it on for every job under Censored at dash.wyzie.io, or per request with music=1 or music=0.
Two voices at once. When one person swears while another is talking, speech recognition follows only one of them. With music=1, Wyzie also finds the moments where two people talk at the same time, splits the two voices and listens to the quieter one too. This part is 640 of the 1,315. For every minute with no overlap to listen to, 600 comes back, so a file with none (overlap: not_needed) keeps only 40 a minute. All 640 come back when the pass fails (overlap: failed), and the 600 when that audio was already recognised. Never more than 640 a minute comes back. The job result says overlap: used, not_needed or failed, and so does the x-wyzie-overlap header.
For example, a 10-minute episode with music=1 is charged 19,250 when it starts. If the gaps between words are quiet and nobody talks over anyone, 6,750 + 6,000 = 12,750 comes back, and the job costs 6,500.
POST /censor#
Send a video or audio file as the request body, or as the multipart field media, with your key in the Authorization header. An upload can be up to about 100 MB. For a longer video, send only the audio track or use POST /censor/url.
Over 100 MB, or on a connection that drops, send the file in chunks with a resumable upload and start the job with its upload_id.
curl -X POST "https://sub.wyzie.io/censor?mode=bleep&audio_track=fr" \
-H "Authorization: Bearer YOUR_KEY" \
-H "Content-Type: video/mp4" --data-binary @movie.mp4Send your key in the Authorization: Bearer YOUR_KEY header (or X-Wyzie-Key: YOUR_KEY) on every Censored call. ?key= still works for older clients and for links you paste into a browser or player, but a key in an address can end up in logs and browser history.
| Parameter | Example | Description |
|---|---|---|
| key | YOUR_KEY | A Pro API key. Required. |
| language | en | The spoken language as a two-letter code. Leave it out to detect it. |
| audio_track | 1, fr | The audio track to censor: its number (0 is the first and the default) or its language, such as fr. Every other track is kept unchanged. |
| mode | bleep | bleep, mute or sound, for this request only. |
| sound | chime | chime, soft, static or whoosh, used with mode sound. |
| level | strict | lenient, moderate, strict or super_strict. Uses that level's lists instead of the account lists for this request. Also kids, teen or family. |
| lists | profanity_strong,slurs | Comma-separated list names, used instead of the account lists for this request. |
| words | rape,=ass | Comma-separated words, added to your blocked words for this request. The same rules apply, so =word is exact. |
| allow | hell | Comma-separated words never to censor, added to your Never Censor list for this request. |
| pad_ms | 60 | Padding in milliseconds, 0 to 300. |
| music | 1 | 1 turns on loud music mode for this request (1,315 more requests a minute, and the parts a file doesn't need come back), 0 turns it off. Leave it out to use the account setting. |
| cache | 0 | 0 keeps this audio out of the recognition cache. |
| profile | Kids | One of the account's named profiles. See Named Profiles below. |
| review | 1 | 1 stops after finding the words so you can check the list first. See Review Before Rendering below. |
| ranges_file | @film.edl | An EDL, VideoSkip .skp or JSON filter file whose ranges are silenced. No speech recognition runs. See below. |
| imdb, tmdb | tt0111161, 603 | The film or show this is, as an IMDb id or a TMDB number. That title's Never Censor words from your account apply to this request. |
| webhook | https://example.com/wyzie | An https URL to call when the job ends. See Webhooks. |
Send level or lists, not both. Both together, or a level that does not exist, get a 400 before anything is charged. The details text is one of these:
Send level or lists, not both.
level is lenient, moderate, strict, super_strict, kids, teen, family.A short file is answered within about 50 seconds with the finished job. A longer one answers 202 with an id and a poll link: ask GET /censor/:id every few seconds until it is done.
{
"id": "c3f1a9e2b7d84c10",
"cost": 1220,
"charged": 1220,
"duration": 95,
"status": "processing",
"queue": 1,
"poll": "https://sub.wyzie.io/censor/c3f1a9e2b7d84c10"
}Apply a filter file. Send ranges_file (a multipart file or its text; on POST /censor/url, a string in the JSON body) and Wyzie silences exactly its ranges instead of listening for words. It reads the files GET /censor/:id/filter makes: an EDL, a VideoSkip .skp or the JSON filter, up to 2 MB and 20,000 ranges. No speech recognition runs, so it costs only the render, 10 requests per started minute. It can't be combined with music or review: that is a 400 (Invalid ranges_file), and nothing is charged. A broken line is a 400 that names the line.
curl -X POST "https://sub.wyzie.io/censor" \
-H "Authorization: Bearer YOUR_KEY" \
-F "[email protected]" -F "[email protected]"Words allowed for one title. Your account can keep Never Censor words for single films and shows, up to 100 titles with 50 words each, saved as censor.titles in the account's Censored settings. Name the title with imdb=tt0111161 or tmdb=603 on POST /censor, POST /censor/url or POST /censor/mask, and that title's words are never muted. A bad id is a 400: imdb is an IMDb id like tt0111161, tmdb is a TMDB number like 603. Edit them in the console at dash.wyzie.io: your account, then Wyzie Censored, then Never Censor by Title. Add a title by its IMDb id, TMDB number or a link to its page, then its words.
Retrying safely. Send an Idempotency-Key header (1 to 100 letters, digits or _.:-) on the calls that start a job: POST /censor, POST /censor/url, POST /censor/batch and POST /censor/mask. Send the same request again with the same key, for example after a timeout, and you get the first answer back with Idempotent-Replayed: true, with no second job and no second charge. A duplicate while the first is still running gets 409: try again after 5 seconds. A successful answer is kept for 24 hours, and an error answer frees the key, so a corrected retry runs. The key belongs to your API key and the route. The same key with a different request is a 422 (idempotency_key_reused), so use a new key for each new request. The SDKs send one for you and reuse it when they retry.
curl -X POST "https://sub.wyzie.io/censor/url" \
-H "Authorization: Bearer YOUR_KEY" \
-H "Idempotency-Key: 7f3c2a90-film-night" \
-H "Content-Type: application/json" \
-d '{"url":"https://example.com/film.mp4","imdb":"tt0111161"}'The result keeps the video stream exactly as it was and re-encodes only the audio (AAC, 160 kb/s). MP4, MOV, M4V and WebM come back as MP4, other video as MKV, and audio on its own as M4A.
Every audio track, subtitle track and chapter is kept. Only one audio track is censored: the first, or the one audio_track names. The result's audio_track says which one. An MP4 can't hold every format, so tracks it can't hold are left out and listed in dropped, such as audio 2 (dts, fre). A track that isn't in the file is a 400 with reason: "audio_track_not_found", and nothing is charged.
{
"status": "done",
"audio_track": 1,
"dropped": ["audio 2 (dts, fre)"],
"warnings": [
"multiple_audio_tracks: 3 audio tracks (eng, fre, fre). Only track 1 (fre) is censored; the others are copied unchanged. Choose with audio_track."
]
}Checks Before the Charge#
Before anything is charged, Wyzie checks the file. A file it can't work with is refused with 400, 413 or 422, charged: false and a reason: unreadable, drm, no_audio, unsupported_audio_codec, bad_duration, too_long, bad_audio, too_many_streams, silent, music_only or audio_track_not_found. The details say how to fix it.
{
"code": 422,
"message": "The audio is silent",
"details": "No sound was found in this file's audio. Nothing was charged.",
"reason": "silent",
"charged": false,
"docs": "https://docs.wyzie.io/subs/usage/censored"
}A job that goes ahead can carry warnings, in the 202 answer and in the result. Each one is name: text, and the names are multiple_audio_tracks, surround, variable_frame_rate, quiet_audio and dropped_tracks. They change nothing about the job; they tell you what to check.
POST /censor/url#
Send a JSON body with key and url, and the server fetches the media itself. It works with a direct video or audio file, an HLS or DASH stream, and video pages on many sites that allow server downloads. Media can be up to 4 hours and 3 GB, and video is fetched at up to 720p. The other options are the same as for POST /censor, and dryRun: true returns the length and the price without charging anything. It answers 202 with an id to poll. The price is set from the length the link reports. If the file turns out longer, the extra minutes are charged before any work starts, or the job stops and is refunded in full when your balance can’t cover them.
curl -X POST "https://sub.wyzie.io/censor/url" \
-H "Authorization: Bearer YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{"url":"https://example.com/film.mp4","mode":"mute","audio_track":0}'curl -X POST "https://sub.wyzie.io/censor/url" \n -H "Authorization: Bearer YOUR_KEY" \n -H "Content-Type: application/json" \n -d '{"url":"https://example.com/ep1.mp4","dryRun":true}'{ "duration": 1320, "title": "Episode 1", "estimateRequests": 13420 }Some sites, YouTube among them, block downloads from servers. Those links are refused with a 422 before anything is charged. To clean videos on those sites, use Censored Live in the browser extension, which works from the captions while you watch.
POST /censor/batch#
Clean up to 20 links in one call (Pro). Send JSON with key, items (1 to 20 objects, each with a url and any options of its own) and any option of POST /censor/url for every item: language, mode, sound, level, lists, words, allow, pad_ms, music, profile, review and cache. An item’s own options win. A batch takes links only: uploads can’t be batched, because the server keeps no stored file a batch could point at.
curl -X POST "https://sub.wyzie.io/censor/batch" -H "Authorization: Bearer YOUR_KEY" -H "Content-Type: application/json" -d '{"mode":"mute","items":[{"url":"https://example.com/ep1.mp4"},{"url":"https://example.com/ep2.mp4","level":"kids"}]}'{
"id": "b7e41c09-5d2a-4f3b-9c8e-1a6d0f2b3c45",
"status": "running",
"poll": "https://sub.wyzie.io/censor/batch/b7e41c09-5d2a-4f3b-9c8e-1a6d0f2b3c45"
}It answers 202 with an id, a status and a poll link. Each item is its own POST /censor/url job, at the same price as a single job, charged when it starts. A key runs 2 items at a time. X-Wyzie-Cost is the cost of the items the call itself started.
The first refused charge (402) stops the batch: that item and every item after it end skipped_no_balance, uncharged, and the batch ends with stopped: no_balance. Any other refusal fails that item and the rest, and the batch ends with stopped: refused. A failed item is refunded by its own job. An unsupported link fails with 422 and is not charged.
With dryRun: true, each item is probed and priced and nothing is charged. The answer is 200 with items (index, url, supported, duration and estimateRequests) and the total estimateRequests. Media over 4 hours is supported: false with an estimate of 0.
{
"items": [
{
"index": 0,
"url": "https://example.com/ep1.mp4",
"supported": true,
"duration": 1320,
"estimateRequests": 13420
}
],
"estimateRequests": 13420
}Add webhook and Wyzie calls it once, with the event batch.done, when every item has ended. The reference is censor-batch:<id>, and the payload has kind: "censor", the counts, cost and charged.
GET /censor/batch/:id#
Poll it with the key that started it. It is free and kept 24 hours. The answer has id, status (running, done or stopped), stopped (no_balance or refused), count, counts, cost, charged, poll and items. Each item has its status (pending, running, done, failed or skipped_no_balance), its job id and poll link, media once done, cost and charged.
curl "https://sub.wyzie.io/censor/batch/b7e41c09-5d2a-4f3b-9c8e-1a6d0f2b3c45" -H "Authorization: Bearer YOUR_KEY"{
"id": "b7e41c09-5d2a-4f3b-9c8e-1a6d0f2b3c45",
"status": "stopped",
"stopped": "no_balance",
"count": 2,
"counts": {
"pending": 0,
"starting": 0,
"running": 0,
"done": 1,
"failed": 0,
"skipped_no_balance": 1
},
"cost": 12200,
"charged": 12200,
"createdAt": "2026-11-12T13:00:00.000Z",
"poll": "https://sub.wyzie.io/censor/batch/b7e41c09-5d2a-4f3b-9c8e-1a6d0f2b3c45",
"items": [
{
"index": 0,
"url": "https://example.com/ep1.mp4",
"status": "done",
"id": "3f0c9a52-8a3e-4c7e-9d0b-6a1f2e4b7c10",
"poll": "https://sub.wyzie.io/censor/3f0c9a52-8a3e-4c7e-9d0b-6a1f2e4b7c10",
"media": "https://sub.wyzie.io/censor/3f0c9a52-8a3e-4c7e-9d0b-6a1f2e4b7c10/media?e=1794480000&t=...",
"cost": 12200,
"charged": 12200
},
{
"index": 1,
"url": "https://example.com/ep2.mp4",
"status": "skipped_no_balance",
"cost": 0
}
]
}The POST can answer 400, 401, 402, 403, 413 and 503. The GET can answer 401 and 404.
GET /censor/:id#
Pass the same key that started the job. While it runs you get 202 with status: processing. When it is done you get the result: where each word was hidden, how many there were, what it cost, and a media link to the clean file. Only the key that started a job can read it, and results are kept for an hour.
{
"id": "3f0c9a52-8a3e-4c7e-9d0b-6a1f2e4b7c10",
"status": "done",
"cost": 229075,
"charged": 229075,
"duration": 7134,
"count": 37,
"music": "used",
"overlap": "used",
"ranges": [
{ "start": 312.48, "end": 312.91, "category": "profanity_strong" }
],
"media": "https://sub.wyzie.io/censor/3f0c9a52-8a3e-4c7e-9d0b-6a1f2e4b7c10/media?e=1794480000&t=...",
"filter": "https://sub.wyzie.io/censor/3f0c9a52-8a3e-4c7e-9d0b-6a1f2e4b7c10/filter",
"subtitle": "https://sub.wyzie.io/censor/3f0c9a52-8a3e-4c7e-9d0b-6a1f2e4b7c10/subtitle",
"ledger": { "charged": 229075, "refunded": 0, "net": 229075, "reason": null },
"expiresAt": "2026-11-12T15:00:00.000Z"
}ledger shows what the job cost: charged (everything taken for it), refunded (what came back, after a failure, an unused part or a cache hit), net (what it cost in the end) and reason (why it failed, refund pending: ... while a refund is still on its way, or null).
The Media Link#
The media link is signed and carries no API key, so you can put it straight into a <video> tag or hand it to a player. It supports seeking and works from any website. It works for 15 minutes and 20 full downloads, so fetch the file right away. The file itself is kept for 1 hour: call GET /censor/:id again with your key for a fresh link. An expired link answers 403 (Link expired or not valid), and one used 20 times answers 429 (Download limit reached).
A finished job also has filter and subtitle links. Call them with the same key. They cost nothing and work for as long as the job is kept (1 hour).
GET /censor/:id/filter#
The censored stretches as a filter file, so a player can mute them while it plays the original video. No new file is made. format picks the file:
| Format | Meaning |
|---|---|
json | The default: count and ranges, each with start and end in seconds, action: "mute", category and word. |
edl | One start end 1 line per range, in seconds (1 means mute). Many media players read it when it sits next to the video with the same name. |
skp | A skip file: each range as a time line, then a mute rule with its category and how strong the word is. |
curl -o film.edl "https://sub.wyzie.io/censor/3f0c9a52-8a3e-4c7e-9d0b-6a1f2e4b7c10/filter?format=edl" -H "Authorization: Bearer YOUR_KEY"{
"count": 1,
"ranges": [
{
"start": 312.48,
"end": 312.91,
"action": "mute",
"category": "profanity_strong",
"word": "f******"
}
]
}word is what was said, masked after its first letter (f***). The answer never shows the words on your lists.
GET /censor/:id/subtitle#
What was said in the file, as a subtitle with the censored words masked the same way. format is srt (the default), vtt or txt (plain text, one cue a line). style=bleep writes [bleep] in place of each hidden word instead of f*** ([muted] when the job used mode mute). Add marks=1 to also show [bleep] on screen while a word is hidden ([muted] when the job used mode mute). Jobs made before 14 November 2026 have no transcript and answer 404.
curl -o film.clean.srt "https://sub.wyzie.io/censor/3f0c9a52-8a3e-4c7e-9d0b-6a1f2e4b7c10/subtitle?format=srt&marks=1" -H "Authorization: Bearer YOUR_KEY"
curl -o film.clean.txt "https://sub.wyzie.io/censor/3f0c9a52-8a3e-4c7e-9d0b-6a1f2e4b7c10/subtitle?format=txt&style=bleep" -H "Authorization: Bearer YOUR_KEY"Both answer 401 without a key, 404 for another key's job or one past its hour, and 409 while the job runs or after it failed.
Review Before Rendering#
Add review=1 to POST /censor or POST /censor/url to check what was found before anything is hidden. The job is charged as usual, finds the words and stops. GET /censor/:id then answers with status: review, the hits, an edit link and a render link. The webhook fires when the review is ready. When the job store is down, review=1 answers 503 before anything is charged.
curl -X POST "https://sub.wyzie.io/censor/url" -H "Authorization: Bearer YOUR_KEY" -H "Content-Type: application/json" -d '{"url":"https://example.com/film.mp4","review":"1"}'{
"id": "3f0c9a52-8a3e-4c7e-9d0b-6a1f2e4b7c10",
"status": "review",
"hits": [
{
"id": "h1",
"word": "f******",
"start": 312.48,
"end": 312.91,
"category": "profanity_strong",
"confidence": 1,
"keep": true
},
{
"id": "h2",
"word": "h***",
"start": 640.1,
"end": 640.42,
"category": "profanity_mild",
"confidence": 0.6,
"keep": true
}
],
"edit": "https://sub.wyzie.io/censor/3f0c9a52-8a3e-4c7e-9d0b-6a1f2e4b7c10/hits",
"render": "https://sub.wyzie.io/censor/3f0c9a52-8a3e-4c7e-9d0b-6a1f2e4b7c10/render",
"expiresAt": "2026-11-12T15:00:00.000Z"
}Each hit has an id, the word masked after its first letter (f***), start and end in seconds, the category, a confidence and keep. Ranges you add have added: true and the category added. confidence is 1 when the word was heard in the mix, and 0.6 when it was heard only in the separated voices of loud music mode (music=1).
A review=1 job is kept until expiresAt (24 hours). Until it is rendered, the media link answers 404 and the filter and subtitle exports answer 409. The rendered file is then kept for 1 hour.
GET /censor/:id/transcript#
Free, while a review=1 job is in review, and only for the key that started it: every word of the transcript with its place i, the word w, and s and e in seconds. A word a hit covers arrives masked and carries the hit's id in hit. Use the places to hide any word with PATCH /censor/:id/hits. A job without review=1, or one already rendered, answers 404.
curl "https://sub.wyzie.io/censor/3f0c9a52-8a3e-4c7e-9d0b-6a1f2e4b7c10/transcript" -H "Authorization: Bearer YOUR_KEY"{
"id": "3f0c9a52-8a3e-4c7e-9d0b-6a1f2e4b7c10",
"state": "review",
"count": 9841,
"words": [
{ "i": 0, "w": "Oh", "s": 312.1, "e": 312.3 },
{ "i": 1, "w": "f******", "s": 312.48, "e": 312.91, "hit": "h1" }
]
}PATCH /censor/:id/hits#
Send JSON with keep and drop (lists of hit ids) and add (ranges with start and end in seconds, at most 500, muted exactly as sent). It is free and answers the job, still in review. A bad id or range is a 400, and a job that is not in review answers 409.
curl -X PATCH "https://sub.wyzie.io/censor/3f0c9a52-8a3e-4c7e-9d0b-6a1f2e4b7c10/hits" -H "Authorization: Bearer YOUR_KEY" -H "Content-Type: application/json" -d '{"drop":["h2"],"add":[{"start":12.5,"end":13},{"word":57},{"from":120,"to":123}]}'An added range can also name words by their place in GET /censor/:id/transcript: {"word": 57} for one word, or {"from": 120, "to": 123} for a run of words (both ends included). A place that is not in the transcript is a 400.
POST /censor/:id/render#
Renders the edited list: kept hits are hidden the way the job would have hidden them, and added ranges exactly as sent. It answers 202 with poll and cost: 0, because the job already paid. Then poll GET /censor/:id as usual. A job renders once (409 after that). When the render fails, the job goes back to review with an error.
curl -X POST "https://sub.wyzie.io/censor/3f0c9a52-8a3e-4c7e-9d0b-6a1f2e4b7c10/render" -H "Authorization: Bearer YOUR_KEY"POST /censor/mask#
Censored Live is for players that already know when each word is said, from captions or a subtitle. You send that timing, Wyzie answers with the stretches to hide, and your player mutes or bleeps them itself while the video plays. There is no upload and no new file, which is why it costs only 1 request per started minute of the time the timing covers.
Send a JSON body with key and either words (each with word, start and end) or cues (each with start, end and text). Times are in seconds. Word times are used as they are. With cues, Wyzie estimates where each word falls in the line and pads it by another 150 ms. Add duration to keep the ranges inside the video, add for extra blocked words, allow for words to keep, and mode, sound, level, lists or pad_ms to override the account. The body can be up to 2 MB.
curl -X POST "https://sub.wyzie.io/censor/mask" \
-H "Authorization: Bearer YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{"duration":600,"lists":"profanity_mild","cues":[{"start":12.0,"end":14.5,"text":"Oh hell, not again."}]}'{
"ranges": [
{
"start": 12.08,
"end": 12.62,
"category": "profanity_mild",
"word": "h***"
}
],
"mode": "bleep",
"sound": "chime",
"pad_ms": 60,
"count": 1,
"cost": 10
}The answer lists the ranges with their category, plus the mode, sound and padding to play them with. It never returns the words on your lists. The same body from the same key within 6 hours is answered from a cache with cached: true and costs nothing.
Add sdh_cues to also hide the sounds the captions announce in brackets, such as [screaming] or (gunshots): 1 for every kind, or a list of gunfire, screaming, moaning and explicit_music. It is off by default and costs nothing extra. Each one comes back in ranges with the category sdh_cues, its span, word (the kind, such as [screaming]) and mask (the cue and the words it covers). Only annotations in brackets are read, never speech, mostly in English (a few in Spanish, French, German and Portuguese). Any other value is a 400 (Invalid option), and nothing is charged.
curl -X POST "https://sub.wyzie.io/censor/mask" \
-H "Authorization: Bearer YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{"duration":600,"sdh_cues":"screaming,gunfire","cues":[{"start":40.0,"end":42.0,"text":"[screaming]"}]}'{
"ranges": [
{
"start": 40,
"end": 42,
"category": "sdh_cues",
"word": "[screaming]",
"mask": [[0, 0, 0]]
}
]
}Add format: "edl" or format: "skp" (in the body or the query) to get the ranges as that filter file instead of JSON, at the same price and from the same cache. In JSON, each range also has word, masked like f***.
Each range has a category: the list the word is on (profanity_mild, profanity_strong, profanity_severe, slurs, blasphemy, sexual, minced or insults), or custom for one of your blocked words that is on no list.
GET /censor/settings#
Free. GET /censor/settings returns the settings that apply to that key: whether Censored is on, the lists, how many blocked words and Never Censor words there are, the mode, the sound, the padding and the current prices. Players use it to show the right price before they charge anything.
The answer also has level, the level your lists make (or custom when they match none), and available.levels, the lists behind each level. custom and allow are counts, never the words. prices has the current prices in requests per started minute: censor, censorCacheHit, censorLive and censorMusic.
{
"enabled": true,
"account": true,
"lists": ["profanity_strong", "profanity_severe", "slurs"],
"level": "moderate",
"custom": 1,
"allow": 1,
"mode": "bleep",
"sound": "chime",
"pad_ms": 60,
"music": false,
"profiles": [
{
"name": "Kids",
"level": "kids",
"lists": ["...the same as super_strict"],
"words": 3,
"allow": 0
}
],
"available": {
"languages": ["es", "fr", "de", "pt", "...23 codes"],
"lists": [
"profanity_mild",
"profanity_strong",
"profanity_severe",
"slurs",
"blasphemy",
"sexual",
"minced",
"insults",
"mild_damn_hell",
"mild_crude",
"blasphemy_exclamations",
"blasphemy_curses"
],
"modes": ["bleep", "mute", "sound"],
"sounds": ["chime", "soft", "static", "whoosh"],
"levels": {
"lenient": ["profanity_severe", "slurs"],
"moderate": ["profanity_strong", "profanity_severe", "slurs"],
"strict": [
"profanity_mild",
"profanity_strong",
"profanity_severe",
"slurs",
"blasphemy",
"sexual"
],
"super_strict": [
"profanity_mild",
"profanity_strong",
"profanity_severe",
"slurs",
"blasphemy",
"sexual",
"minced",
"insults"
],
"kids": ["...the same as super_strict"],
"teen": [
"profanity_strong",
"profanity_severe",
"slurs",
"sexual",
"blasphemy_curses"
],
"family": [
"mild_crude",
"profanity_strong",
"profanity_severe",
"slurs",
"sexual",
"blasphemy_curses"
]
}
},
"prices": {
"unit": "requests per started minute of media",
"censor": 610,
"censorCacheHit": 10,
"censorLive": 1,
"censorMusic": 1315
}
}Censored Subtitles#
Add censor=1 to a subtitle download link and the words on your account lists are masked in the text, keeping the first letter (f***). It is free. The X-Censor header says on when the words were masked, off when Censored is off for the account (the subtitle comes back unmasked), and ignored for a key with no account. The Stremio add-on, the Bazarr provider and the Jellyfin and Emby plugins use it for clean subtitles.
Add level to the same link to use another level for that one download.
Add lang=es next to censor=1 to mask with that language's lists (a bad code is a 400 Invalid lang). Download links from /search already end in &lang= with the subtitle's base language, so censor=1 on a search link picks the right lists by itself.
curl "https://sub.wyzie.io/c/198e0c4d/id/1955024019?format=srt&encoding=UTF-8&tok=...&lang=es&censor=1"Pricing#
| Use | Requests per started minute |
|---|---|
| A file, uploaded or from a link | 610 |
| The same audio again (recognition reuse) | 10 |
A filter file (ranges_file, no recognition) | 10 |
Loud music mode (music=1), added | +1,315 |
Censored Live (POST /censor/mask) | 1 |
Censored subtitles (censor=1) | Free |
| Filter file and clean subtitle of a finished job | Free |
Rendering a reviewed job (POST /censor/:id/render) | Free |
Minutes are counted from the length of the media, rounded up, with at least 1. A 2-hour film costs 73,200 requests and a 20-minute episode 12,200. If the same audio was recognised before, 600 per minute is refunded as soon as the job knows, so only the 10 for making the file is left. Anything refused before the job starts is free.
Webhooks#
Add webhook with an https URL and Wyzie calls it when the job ends, with the event censor.done or censor.failed. See Webhooks for the signature and retries.
Errors#
| Status | Meaning |
|---|---|
| 400 | A bad option or link. The message says which, and nothing is charged. |
| 401 | No key. |
| 402 | Not enough requests left for this file's length. |
| 403 | A free or invalid key, or Censored is off for the account (code: PRODUCT_OFF, with a link to turn it on). |
| 404 | No job with this id for this key, or the result is more than an hour old. |
| 409 | The job is in the wrong state: still running, waiting in review, not in review for an edit, or already rendered. |
| 413 | The upload is too big, or the media is longer than 4 hours. |
| 422 | The link is on a site that blocks server downloads. Use the browser extension there. |
| 429 | Too many jobs from this key this hour. |
In the SDKs#
The clients do this for you: wyzie-lib 2.24.0 (WyzieError.reason, MediaLinkExpiredError), wyzie-go v1.10.0 (APIError.Reason(), ErrLinkExpired), the Python client 0.10.0 (WyzieError.reason, LinkExpiredError) and wyzie-mcp 0.10.0. They send the key in the header, take audio_track, return warnings and dropped, and ask the job once for a fresh link when a download link has expired.
import {
censor,
censorUrl,
waitCensor,
censorMask,
censorFilter,
censorSubtitle,
censorTranscript,
censorEditHits,
censorRender,
censorBatch,
getCensorBatch,
censorSettings,
} from 'wyzie-lib';
const job = await censor(videoBlob, {
key: 'YOUR_KEY',
level: 'strict',
mode: 'bleep',
}); // waits; over 100 MB it uploads in chunks
const { estimateRequests } = await censorUrl('https://example.com/ep1.mp4', {
key: 'YOUR_KEY',
dryRun: true,
}); // free
const started = await censorUrl('https://example.com/ep1.mp4', {
key: 'YOUR_KEY',
mode: 'mute',
wait: false,
});
const done = await waitCensor(started.id, { key: 'YOUR_KEY' });
const edl = await censorFilter(done.id, { key: 'YOUR_KEY', format: 'edl' });
const srt = await censorSubtitle(done.id, {
key: 'YOUR_KEY',
format: 'srt',
marks: true,
});
const live = await censorMask({
key: 'YOUR_KEY',
cues: [{ start: 12, end: 14.5, text: 'Oh hell, not again.' }],
});
// Review first: stop at the hit list, edit it, then render (free)
const review = await censor(videoBlob, { key: 'YOUR_KEY', review: true });
const words = await censorTranscript(review.id, { key: 'YOUR_KEY' });
await censorEditHits(
review.id,
{ drop: ['h2'], add: [{ start: 12.5, end: 13 }] },
{ key: 'YOUR_KEY' },
);
await censorRender(review.id, { key: 'YOUR_KEY' });
const batch = await censorBatch({
key: 'YOUR_KEY',
mode: 'mute',
items: [{ url: 'https://example.com/ep1.mp4' }],
});
const progress = await getCensorBatch(batch.id, 'YOUR_KEY');
const settings = await censorSettings('YOUR_KEY'); // free: level, lists, pricesc := wyzie.New("YOUR_KEY")
f, _ := os.Open("movie.mp4")
job, err := c.Censor(ctx, f, wyzie.CensorOptions{Level: wyzie.LevelStrict, Mode: "bleep"})
est, err := c.EstimateCensorURL(ctx, "https://example.com/ep1.mp4", wyzie.CensorOptions{}) // free
started, err := c.CensorURL(ctx, "https://example.com/ep1.mp4", wyzie.CensorOptions{Mode: "mute"})
done, err := c.WaitCensor(ctx, started.ID, wyzie.JobWait{})
edl, err := c.CensorFilter(ctx, done.ID, "edl")
srt, err := c.CensorSubtitle(ctx, done.ID, "srt", true)
live, err := c.CensorMask(ctx, wyzie.CensorMaskRequest{Cues: []wyzie.Cue{{Start: 12, End: 14.5, Text: "Oh hell, not again."}}})
words, err := c.CensorTranscript(ctx, reviewID) // a job started with Review: true
_, err = c.CensorEditHits(ctx, reviewID, wyzie.CensorHitsEdit{Drop: []string{"h2"}, Add: []wyzie.CensorSpan{{Start: 12.5, End: 13}}})
_, err = c.CensorRender(ctx, reviewID)
batch, err := c.CensorBatch(ctx, wyzie.CensorBatchRequest{Items: []wyzie.CensorBatchItem{{URL: "https://example.com/ep1.mp4"}}})
progress, err := c.GetCensorBatch(ctx, batch.ID)
settings, err := c.CensorSettings(ctx) // free: level, lists, pricesfrom wyzie_subs import Wyzie
wyzie = Wyzie("YOUR_KEY")
job = wyzie.censor("movie.mp4", level="strict", mode="bleep") # waits; over 100 MB it uploads in chunks
est = wyzie.censor_url("https://example.com/ep1.mp4", dry_run=True)["estimateRequests"] # free
started = wyzie.censor_url("https://example.com/ep1.mp4", mode="mute", wait=False)
done = wyzie.wait_censor(started["id"])
open("movie.edl", "w").write(wyzie.censor_filter(done["id"], format="edl"))
srt = wyzie.censor_subtitle(done["id"], format="srt", marks=True)
live = wyzie.censor_mask(cues=[{"start": 12, "end": 14.5, "text": "Oh hell, not again."}])
review = wyzie.censor("movie.mp4", review=True) # stops at the hit list
words = wyzie.censor_transcript(review["id"])
wyzie.censor_edit_hits(review["id"], drop=["h2"], add=[{"start": 12.5, "end": 13}])
wyzie.censor_render(review["id"]) # free
batch = wyzie.censor_batch([{"url": "https://example.com/ep1.mp4"}], mode="mute")
progress = wyzie.get_censor_batch(batch["id"])
settings = wyzie.censor_settings() # free: level, lists, priceswyzie-mcp tools: censor_video_url, censor_video_file, censor_batch, censor_batch_status,
censor_mask, censor_settings, censor_filter, censor_subtitle, censor_transcript,
censor_edit_hits, censor_render, job_status (kind censor).
Without confirm: true a paid tool only answers the price; call it again with confirm: true.
{ "tool": "censor_video_url", "arguments": { "url": "https://example.com/ep1.mp4", "level": "strict", "confirm": true } }Examples#
# Clean a short clip (answered within ~50 s) and save the result
curl -s -X POST "https://sub.wyzie.io/censor?mode=sound&sound=soft" \
-H "Authorization: Bearer YOUR_KEY" \
-H "Content-Type: video/mp4" --data-binary @clip.mp4 > job.json
# The media link works for 15 minutes and 20 downloads: fetch it right away
curl -L "$(jq -r .media job.json)" -o clip-clean.mp4From JavaScript, start a job and wait for the clean file:
const auth = { Authorization: 'Bearer YOUR_KEY' };
const res = await fetch('https://sub.wyzie.io/censor/url', {
method: 'POST',
headers: { ...auth, 'Content-Type': 'application/json' },
body: JSON.stringify({ url: 'https://example.com/film.mp4' }),
});
let job = await res.json();
while (job.status !== 'done') {
if (job.status === 'failed' || job.code) throw new Error(job.details);
await new Promise((r) => setTimeout(r, 10_000));
job = await fetch(`https://sub.wyzie.io/censor/${job.id}`, {
headers: auth,
}).then((r) => r.json());
}
// signed, plays and seeks for 15 minutes (20 downloads);
// ask GET /censor/:id again within the hour for a fresh link
video.src = job.media;