SilentAPI

Media and OCR

Image tools, file tools, OCR, moderation, music, and parameter handling

File endpoints accept a multipart file upload or a JSON data URL. Moderation accepts text or chat messages. Music accepts a prompt and an optional image. Authenticate with your SilentAPI API key.

Response values below are examples, not current prices. Base64 response strings show only part of the output. The cost field is in credits. See the [ Services ] page or [ Calculator ] for current prices.

For JSON file requests, set file to a complete data URL. The examples use a variable with a small PNG:

IMAGE_DATA='data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mP8/x8AAwMCAO+aBZkAAAAASUVORK5CYII='

Image Compress

POST /v1/images/compress creates a smaller image and removes metadata by default.

curl https://api.silentapi.com/v1/images/compress \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "file=@photo.png" \
  -F "format=webp" \
  -F "quality=75" \
  -F "response_format=b64_json"

curl https://api.silentapi.com/v1/images/compress \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{\"file\":\"$IMAGE_DATA\",\"format\":\"webp\",\"quality\":75,\"response_format\":\"b64_json\"}"

{
  "object": "image",
  "data": [{ "b64_json": "UklGRg==", "mime_type": "image/webp" }],
  "width": 1600,
  "height": 900,
  "bytes_in": 842100,
  "bytes_out": 126400,
  "cost": 0.12
}

The default response is binary. Use -o output.webp with curl to save it. Read X-SilentAPI-Cost for the charge. Compression is charged per call. Input must fit the 20 MB and image dimension limits. See the [ Services ] page for current prices.

Image Convert

POST /v1/images/convert changes an image to a selected format. The request fields and response shape match Image Compress. The format field is required.

curl https://api.silentapi.com/v1/images/convert \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "file=@photo.png" \
  -F "format=jpeg" \
  -F "quality=82" \
  -F "response_format=b64_json"

curl https://api.silentapi.com/v1/images/convert \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{\"file\":\"$IMAGE_DATA\",\"format\":\"jpeg\",\"quality\":70,\"response_format\":\"b64_json\"}"

{
  "object": "image",
  "data": [{ "b64_json": "/9j/", "mime_type": "image/jpeg" }],
  "width": 1200,
  "height": 800,
  "bytes_in": 520000,
  "bytes_out": 98000,
  "cost": 0.12
}

Conversion is charged per call. Input must fit the 20 MB and image dimension limits. See the [ Services ] page for current prices.

Remove File Metadata

POST /v1/files/metadata/remove removes metadata from images, PDF files, and Office files.

curl https://api.silentapi.com/v1/files/metadata/remove \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "file=@document.pdf" \
  -F "response_format=b64_json"

curl https://api.silentapi.com/v1/files/metadata/remove \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{\"file\":\"$IMAGE_DATA\",\"response_format\":\"b64_json\"}"

{
  "object": "file",
  "data": [{ "b64_json": "JVBERi0=", "mime_type": "application/pdf" }],
  "removed": ["document-properties", "xmp"],
  "cost": 0.02
}

Input is limited to 20 MB. Encrypted PDF files are not supported. JPEG orientation and image color profiles can remain. HEIC input becomes JPEG. Office processing removes document properties and some author fields. It does not remove document content.

Metadata removal is charged per call. A file conversion can add a transform charge. See the [ Services ] page for current prices.

OCR

POST /v1/ocr reads text from images and documents. Set model to one of these values:

curl https://api.silentapi.com/v1/ocr \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "file=@scan.pdf" \
  -F "model=silentapi-ocr" \
  -F "outputFormat=blocks" \
  -F "pages=1-3"

curl https://api.silentapi.com/v1/ocr \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{\"file\":\"$IMAGE_DATA\",\"model\":\"silentapi-ocr\",\"outputFormat\":\"text\"}"

OCR

OCR supports images and PDF files. It uses English OCR data and allows up to 50 selected pages.

Selected pages must exist. Use output_format as an alternative to outputFormat.

{
  "object": "ocr",
  "model": "silentapi-ocr",
  "pages": [{ "page": 1, "text": "Example text", "source": "ocr" }],
  "usage": { "pages": 1, "ocrPages": 1, "cpuMs": 215 },
  "cost": 0.17
}

The source field shows how SilentAPI read each page:

Blocks can include these fields:

Charges use pages, OCR pages, and image transforms. See the [ Services ] page for current prices.

Google Cloud Vision Document Text Detection

Document Text Detection is Google Cloud Vision's text reader for dense text and documents. On POST /v1/ocr, set model to vision-document-text-detection to get the same OCR response shape as the other models. It reads one image per call (JPEG, PNG, WebP or GIF, up to 20 MB); PDF input is not supported. The older model name google-ocr still works. For the full Google Cloud Vision response, call POST /v1/google/vision-document-text-detection, or vision-text-detection for short text in photos.

curl https://api.silentapi.com/v1/ocr \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "file=@scan.png" \
  -F "model=vision-document-text-detection" \
  -F "outputFormat=blocks"

Mistral OCR 4.1

Mistral OCR 4.1 reads images and documents. Input is limited to 20 MB. Page selection uses numbers from 1 to 1,000. Provider limits also apply.

curl https://api.silentapi.com/v1/ocr \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "file=@scan.pdf" \
  -F "model=mistral-ocr-4-1" \
  -F "pages=1-3" \
  -F "outputFormat=markdown" \
  -F "table_format=markdown"

curl https://api.silentapi.com/v1/ocr \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{\"file\":\"$IMAGE_DATA\",\"model\":\"mistral-ocr-4-1\",\"outputFormat\":\"blocks\",\"include_image_base64\":false}"

{
  "object": "ocr",
  "model": "mistral-ocr-4-1",
  "pages": [{ "page": 1, "markdown": "# Example document", "images": [] }],
  "usage": { "pages": 1 },
  "cost": 0.8
}

Each page has page and markdown. The other fields appear when the response includes them:

A document_annotation field can appear at the top level of the response.

Mistral OCR is charged per OCR page. Annotation options use the annotated page rate. See the [ Services ] page for current prices.

Moderation

POST /v1/moderations checks text or chat messages with mistral-moderation-2603, also named Mistral Moderation 2.

curl https://api.silentapi.com/v1/moderations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"mistral-moderation-2603","input":["Example text to check"]}'

curl https://api.silentapi.com/v1/moderations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "model=mistral-moderation-2603" \
  -F "input=Example text to check"

Moderation does not accept a file upload or data URL as media input. Use input for text, or messages for chat. When both exist, messages is used.

{
  "id": "modr-12345678-1234-4234-8234-123456789012",
  "model": "mistral-moderation-2603",
  "results": [{ "flagged": false, "categories": { "hate": false }, "category_scores": { "hate": 0.01 } }],
  "cost": 0.0203
}

Category names depend on the model response. Successful moderation calls have a call charge and an input token charge. Shared service and account media limits apply. See [ Rate Limits ] and the [ Services ] page for current prices.

Music

POST /v1/audio/music creates music with lyria-3-pro or lyria-3-clip. The default is lyria-3-clip.

curl https://api.silentapi.com/v1/audio/music \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{\"model\":\"lyria-3-clip\",\"prompt\":\"A quiet electronic morning\",\"image\":\"$IMAGE_DATA\",\"response_format\":\"wav\"}"

curl https://api.silentapi.com/v1/audio/music \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "prompt=A quiet electronic morning" \
  -F "model=lyria-3-clip" \
  -F "image=@reference.png"

{
  "object": "music",
  "created": 1770000000,
  "data": [{ "b64_json": "UklGRg==", "mime_type": "audio/wav", "lyrics": null }],
  "silentapi": { "cost": 4, "fallback": null, "model": "lyria-3-clip", "model_name": "Lyria 3 Clip" }
}

Music always returns JSON with base64 audio. The output media type can depend on the route that served the request. Lyrics can be null. There is no binary response option.

The file tool dimension limit does not apply to this endpoint. Music is charged per generation. See the [ Services ] page for current prices.

Parameter Handling

Each model accepts its own set of options. When you send a valid OpenAI style option that the model does not support, SilentAPI changes it instead of failing the request. It converts the output, uses the closest value the model supports, or leaves out an option that has no effect. SilentAPI does not send a value that the model rejects.

Audio responses have the X-SilentAPI-Format header. It shows the real format of the audio, for example wav. The Content-Type header matches it.

Speech

POST /v1/audio/speech accepts response_format values mp3, wav, opus, aac, flac, and pcm on every speech model.

When you do not send response_format, Voxtral returns mp3 and Gemini returns wav. SilentAPI makes WAV files from PCM audio by adding a WAV header. The audio does not change. PCM output is raw signed 16 bit samples, little endian, mono. The Content-Type gives the sample rate, for example audio/pcm;rate=24000;channels=1.

curl https://api.silentapi.com/v1/audio/speech \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"voxtral-mini-tts","input":"Your request is ready.","voice":"en_paul_neutral","response_format":"wav"}' \
  -D headers.txt \
  -o speech.wav

Transcription

POST /v1/audio/transcriptions accepts every response_format below on every speech to text model. SilentAPI builds the requested shape from the model output.

Voxtral Mini Transcribe gives a time for each segment. Gemini 3.5 Transcribe gives segment times for some requests. GPT Transcribe gives text only. When the model gives no times, SilentAPI returns one segment or one subtitle for the full audio. Its end time is the audio length. The length is exact for WAV files and estimated from the file size for other formats.

The text, srt, and vtt formats return no JSON. Read the charge in the X-SilentAPI-Cost header. An unknown response_format returns json. SilentAPI sends language, prompt, and temperature to the model when you send them.

curl https://api.silentapi.com/v1/audio/transcriptions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "file=@recording.mp3" \
  -F "model=voxtral-mini-transcribe" \
  -F "response_format=srt"

Image Generation

POST /v1/images/generations and POST /v1/images/edits change these options to fit the model:

GPT Image models make all n images in one call and change these options:

Music

The X-SilentAPI-Format header and mime_type show the real format. Music models have no length control, so SilentAPI does not send duration or seconds. Lyria 3 Clip makes short clips. For Lyria 3 Pro, write the length you want in the prompt.

Video

Video requests never fail because an option is outside what a model offers. SilentAPI uses the closest supported value:

• duration (or seconds) becomes a whole number of seconds inside the model's range, for example 30 becomes 15 on a model that allows 1 to 15 seconds.
• resolution (or size as WIDTHxHEIGHT) uses the nearest supported resolution, for example 1080p becomes 720p on a model that stops at 720p.
• aspect_ratio uses the nearest supported ratio; a size such as 1080x1920 also sets the ratio.
• The credit hold and the charge use the same adjusted values, so you pay for the video you get.

Embeddings

POST /v1/embeddings accepts encoding_format float (default) or base64 on every model. SilentAPI gets the float vector and makes the base64 string itself. The string holds 32 bit floats, little endian, the same layout as the OpenAI API.

A dimensions value that is equal to or larger than the full size returns the full vector. A value that is not a whole number above 0 is ignored.