Switch Developer API

Build with Switch.

Call Switch’s generation models straight from your code. Every request lands in the same queue Studio uses, with the same receipt and the same tokens.

Create an API key Read the quickstart

Building Switch into your app? Read the developer guidelines

In plain English

One API call. Real images.

You send a prompt and a model. Switch generates the images and drops them straight into your Library, exactly like Studio does. You get a task id back and poll one URL until the pictures are ready. No pipelines to wire up, no servers to run.

What you can build

Four ways in.

Pick the surface that fits how you work. They all hit the same models and the same account.

REST API

Generate from code

POST a prompt, a model, and a count. Images land in your Library and you poll one URL for the result.

MCP

Connect AI assistants

Point Claude, Cursor, or any MCP client at Switch. It can generate, search your library, and check your balance.

Command line

Run it from the terminal

The switch CLI fires the same API from scripts and automations. One login, machine-readable output.

References

Bring your own look

Attach a face or product reference. Your prompt drives the scene, lighting, and camera while the subject stays locked.

See it work

Say it. Get it.

Three everyday calls, start to finish.

One prompt

You write

“a sunset over Miami”

Switch runs

model: switch · count: 4

Four images in your Library, ready to download.

A lookbook

You write

“front · side · three-quarter · over the shoulder”

Switch runs

model: nano_banana_2 · prompts: [ … ]

A four-shot set of the same subject, one prompt each.

With a reference

You write

“a black silk gown on a marble staircase”

Switch runs

model: switch_ultra · references: [ face, body ]

Your subject, kept consistent, in a brand-new scene.

Spend Center

Know what you're spending

Your spend so far, what's left, and where it's going, in real time. Set a daily, weekly, or monthly budget and generation stops before you go over.

Your spend comes with a paid membership.

See memberships

Quickstart

Your first call

Authorize with a Bearer key, POST a model alias, a prompt, and a count. The response gives you a task_id and a status_url you can poll.

model

count

curl -X POST https://www.switchapp.ai/api/v1/tasks \
  -H "Authorization: Bearer $SWITCH_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "switch",
    "prompt": "a sunset over Miami",
    "count": 1
  }'

Authentication

Two kinds of credentials

Switch has two credential types. Which one you use depends on what you are calling, and full access uses both.

API key

sk_switch_live_…

Unlocks the REST API

Image generation and account info over /api/v1/*: the curl examples above, plus the CLI image and whoami commands.

Create one in the API keys section below, then send it as Authorization: Bearer.

OAuth sign-in

browser login

Unlocks the MCP tools

Video, uploads, library search, balance and model listing: the tools exposed over the MCP server.

Used by MCP clients and by switch login --oauth. You sign in through the browser; there is no key to copy.

Rule of thumb: OAuth when your client can open a browser (Claude, Cursor, Codex). A key when it cannot: bots and scripts send the key as Authorization: Bearer to the same MCP endpoint and get every tool. Keys made from this page carry both the REST and the MCP scopes. An OAuth login will not work for the REST API.

Task creation

The request body

POST /api/v1/tasks takes a JSON body. Tasks are scoped to the key’s owner. Count caps at 10 per request.

Field	Type	Required	Default	Notes
model	string (alias)	yes	.	One of the public aliases. See Model aliases.
prompt	string	yes (or prompts)	.	Single prompt repeated count times.
prompts	string[]	yes (or prompt)	.	Explicit per-image prompts. Length sets count automatically.
count	integer	no	1	1 to 10. Ignored if prompts[] is set.
references	object[]	no	[]	{ url: "https://…", type: "face" \| "body" }
folder	string	no	"Generated"	Destination folder name in your Library.
quality	string	no	"2K"	0.5K \| 1K \| 2K \| 4K
scheduled_for	ISO 8601	no	null	The task waits and runs at this time.

API keys

Create and manage keys

Create a key and use it as your Bearer token. Each key is shown once, when you create it. Copy it then, because it can never be displayed again. You can keep up to 25 active keys.

Your keys comes with a paid membership.

See memberships

Revoke

Turn a key off

Click Revoke next to a key and confirm. It stops working immediately, and any further request with it returns:

HTTP/2 401
{ "error": { "code": "INVALID_KEY", "message": "Missing or invalid Switch API key. ..." } }

Revoking is permanent. Create a new key to replace it.

MCP connector

Let an AI assistant use Switch

Switch runs a hosted MCP server, so AI tools like Claude and Cursor can use Switch directly. There is nothing to install and no server to run. You point your client at the endpoint and sign in.

Endpoint

https://www.switchapp.ai/mcp

Connect

Add the endpoint to your MCP client and sign in through the browser when prompted: that is the OAuth step. For step-by-step setup for Claude, ChatGPT, Claude Code, Codex, and Cursor . plus OAuth and troubleshooting, see the MCP setup guide. Then hand your assistant Build a brand. Make a commercial. . one markdown file that teaches any LLM the full Switch workflow: the questions to ask, how to load references, the prompt shape, and how Brand assets saves a reusable brand.

What it gives you

Once connected, the assistant can generate video, upload media, search your library, check your balance, and list available models, all against your own Switch account.

The MCP server is hosted by Switch, so you never run server code. Sign-in uses OAuth, or an API key from this page sent as a Bearer header for bots and scripts that cannot open a browser.

Voice & audio

Give your videos a voice

The same surface the Audio Studio room runs on: save voices, speak with them, build dialogue, and lip sync a video. These routes authenticate with your signed-in Switch session, so the quickest way to try them is the browser console on switchapp.ai. From an AI assistant or a script, connect the MCP connector , its voice tools drive these exact routes on your account.

The voice library, `/api/audio/voices`

One library for every saved voice. Each voice comes back with a kind: reference is an instant voice match saved from a clip (free, unlimited); clone is a trained voice.

GET /api/audio/voicesList your voices, newest first. Query: limit (up to 200), offset, q (name/tag search), kind. Returns voices[] with voiceId, name, kind, ready, tags and a short-lived sampleUrl preview.
POST /api/audio/voicesSave a reference voice from a clip: name plus audioBase64 (with optional format) or a reachable audioUrl, or audioGenerationId to reuse a Library clip. Clip of about 5 to 30 seconds, max 10MB. Free, ready instantly.
PATCH /api/audio/voicesRename or re-tag: speakerId (the voiceId) plus name and/or tags.
DELETE /api/audio/voicesRemove a voice: speakerId. It disappears from every picker.

The routing rule

Reference voices speak through generate; trained voices speak through tts. Concretely: a kind: "reference" voice (ids starting ref_) goes to POST /api/audio/generate as audioRefs: [{ voiceId }] and bills 1 token per second of finished audio. A kind: "clone" voice goes to POST /api/audio/tts as speakerId and bills 1 token per line. Sending a reference voice to tts fails. Through the MCP connector this routing is handled for you, pass the voice’s name.

Speak, `POST /api/audio/tts`

Studio preset voices and trained voices. Body: text (up to 2048 chars) plus voiceType (a preset voice id) or speakerId (a trained voice . wins if both are sent). Delivery knobs: speechRate −50…100, pitch −12…12, loudnessRate −50…100 (0 is normal). format: mp3 or ogg_opus. Returns a playable url (7-day link), durationMs and generationId; the clip lands in your library. Costs 1 token per line.

Generative audio, `POST /api/audio/generate`

The richer sibling: narration, instant voice match, and multi-voice dialogue from one call. Body: text (up to 2048 chars) plus optional references, up to 3 of audioRefs: [{ voiceId | audioUrl | audioData }] (or plain audioRefUrls), addressed in the script as @Audio1 to@Audio3, or a single imageUrl to voice a scene from a picture (never both). Same speechRate / pitch / loudnessRate knobs. Output caps at 120 seconds and bills 1 token per second, reconciled to the real length, a 3-second clip never pays for 60. Returns url, durationSec and generationId.

Examples

// Signed in at www.switchapp.ai — run in the browser console.
const r = await fetch('/api/audio/voices?limit=50', { credentials: 'include' });
const { voices } = await r.json();
// [{ voiceId: 'ref_…', name: 'My narrator', kind: 'reference',
//    ready: true, tags: [], sampleUrl: 'https://…' }, …]
// kind decides the speak route, see "The routing rule" above.
// Optional filters: ?q=narrator  ?kind=reference|clone  ?offset=50

From an AI assistant, the MCP tools

Connected through the MCP connector, four tools cover the whole flow:

voiceYour voice library in one tool, action: list, create (save a reference voice from audio_url or audio_base64 + voice_name), rename (voice_id takes the id or current name, plus new_name), clone (registers a talking-avatar voice, 2 credits), delete.
generate_audioSpeak any script. Pass the saved voice by NAME as voice, reference voices, trained voices and presets are all routed correctly by kind. Or match instantly from clips with reference_audio_url(s) (up to 3, @Audio1 to@Audio3 for dialogue). Knobs: speech_rate −50…100, pitch −12…12, loudness −50…100.
upload_mediaTurn a local or attached clip into a clean URL, pass url or base64 + mime, then feed the returned url into voice create or lip_sync_video as the source.
lip_sync_videoMake the person in a video speak your audio. Simplest path: action: "create" with engine: "best", video_url and sound_file (whole-clip sync). Precise path: identify-face → create with a face and timing → status. Source video mp4/mov, 2 to 60s, up to 100MB, 720p/1080p.

A lip sync render takes about 2 minutes; failed renders refund themselves. The finished video comes back at the source clip’s resolution and lands in your library and feed.

Rate limits

How much, how fast

Per user, image-counted. A count: 10 request uses 10 of your minute budget.

images / minute

1,000

images / day

max count / request

Reference

Command line

Run Switch from your terminal

The switch CLI runs Switch from your terminal, handy for scripts and automation. It calls the same APIs as everything else; there is no second account or backend.

Install

npm install -g @switchapp/cli

Developer preview. The package is not published to npm yet, so this command does not work today; it is how you will install the CLI once it ships.

Log in

# REST: image generation + whoami
switch login --key sk_switch_live_...

# MCP: video, uploads, library
switch login --oauth

See Authentication for which login unlocks which commands. Credentials are stored in ~/.switch/config.json.

Check your account

switch whoami

Generate an image

switch image generate --prompt "a sunset over Miami"
switch image status <task_id>

Troubleshooting

401 / invalid key: image generation needs an Ultra plan key. Create one in the API keys section below.
“unknown tool” on video, upload, or library: those run over MCP; sign in with switch login --oauth first.
402 / out of tokens: top up your account, then retry.

Add --json to any command for machine-readable output.

Usage & budgets

Read spend, set budgets

Read your spend and set budgets from code, the same data the Spend Center shows. Send your API key as Authorization: Bearer.

GET /api/v1/usage

Spend today / this week / this month, your effective limit, remaining, a forecast, and a breakdown by model and by source.

Plain English: how much you've spent, how much is left, and where it went.

GET /api/v1/budget

Your configured budget plus the effective caps actually in force (the most restrictive of your budget, your agency cap, and any admin limit).

Plain English: what your spending limits are right now.

POST /api/v1/budget

Set daily / weekly / monthly caps. Body: { "daily_dollars": 200, "weekly_dollars": null, "monthly_dollars": 1000 }. A blank window means no cap there; a budget can only tighten an agency/admin cap, never loosen it.

Plain English: set the most you want to spend per day, week, or month.

# From the CLI (developer preview)
switch usage              # today's spend, limit, remaining, forecast
switch budget set 200     # cap spending at $200/day
switch budget show        # see your current budget
switch budget disable     # remove your self-set budget

The CLI is in developer preview and not yet published to npm; these commands call the same endpoints above.

API examples

Curl examples

Replace $SWITCH_API_KEY with the key from your API keys section.

curl -X POST https://www.switchapp.ai/api/v1/tasks \
  -H "Authorization: Bearer $SWITCH_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "switch",
    "prompt": "a sunset over Miami",
    "count": 1
  }'

Models

Model aliases

Friendly aliases resolve to the live production model at request time. Pricing is never cached.

Alias	Model	Tokens / image
switch	Switch Model	1
switch_pro	Switch Pro	2
switch_ultra	Switch Ultra	5
nano_banana_2	Nano Banana 2	4
nano_banana_pro	Nano Banana Pro	6
gpt_image	GPT Image 2	3
gpt_image_2	GPT Image 2	3
gpt_image_transparent	GPT Image 2 (transparent)	3
gpt_image_25	GPT Image 2.5	1 / 3 / 8
gpt_image_25_flare	GPT Image 2.5 Flare	1 / 3 / 8

Errors

Error codes

Every error returns JSON shaped like { error: { code, message } }.

Status	Code	Meaning
400	INVALID_JSON	Body is not valid JSON.
400	MODEL_REQUIRED	Body missing the model field.
400	PROMPT_REQUIRED	Provide prompt (string) or prompts (string[]).
400	BAD_COUNT	count must be an integer between 1 and 10.
400	BAD_QUALITY	quality must be 0.5K, 1K, 2K, or 4K.
400	BAD_REFERENCE_TYPE	Reference type must be face or body.
400	BAD_REFERENCE_URL	Reference URL must be valid, public, and http(s).
400	UNKNOWN_ALIAS	Model alias is not in the public set.
400	INACTIVE_MODEL	Alias resolved but production model is no longer active.
400	BAD_SCHEDULE	scheduled_for must be ISO 8601.
401	INVALID_KEY	Missing, malformed, or revoked API key.
403	INSUFFICIENT_SCOPE	Key lacks the tasks:create scope.
403	ULTRA_REQUIRED	Account is not Switch Ultra (or admin).
429	RATE_LIMITED	Per-minute or per-day budget exhausted. Retry-After header included.
500	AUTH_INTEGRITY	Temporary server error. Retry; contact support if it persists.
500	ENQUEUE_FAILED	Could not queue the task. Please retry.
404	TASK_NOT_FOUND	Task status or receipt was not found for this key owner.

Switch Developer API

Build with Switch.

Call Switch’s generation models straight from your code. Every request lands in the same queue Studio uses, with the same receipt and the same tokens.

Create an API key Read the quickstart

Building Switch into your app? Read the developer guidelines

In plain English

One API call. Real images.

What you can build

Four ways in.

Pick the surface that fits how you work. They all hit the same models and the same account.

REST API

Generate from code

POST a prompt, a model, and a count. Images land in your Library and you poll one URL for the result.

MCP

Connect AI assistants

Point Claude, Cursor, or any MCP client at Switch. It can generate, search your library, and check your balance.

Command line

Run it from the terminal

The switch CLI fires the same API from scripts and automations. One login, machine-readable output.

References

Bring your own look

Attach a face or product reference. Your prompt drives the scene, lighting, and camera while the subject stays locked.

See it work

Say it. Get it.

Three everyday calls, start to finish.

One prompt

You write

“a sunset over Miami”

Switch runs

model: switch · count: 4

Four images in your Library, ready to download.

A lookbook

You write

“front · side · three-quarter · over the shoulder”

Switch runs

model: nano_banana_2 · prompts: [ … ]

A four-shot set of the same subject, one prompt each.

With a reference

You write

“a black silk gown on a marble staircase”

Switch runs

model: switch_ultra · references: [ face, body ]

Your subject, kept consistent, in a brand-new scene.

Spend Center

Know what you're spending

Your spend so far, what's left, and where it's going, in real time. Set a daily, weekly, or monthly budget and generation stops before you go over.

Your spend comes with a paid membership.

See memberships

Quickstart

Your first call

Authorize with a Bearer key, POST a model alias, a prompt, and a count. The response gives you a task_id and a status_url you can poll.

model

count

curl -X POST https://www.switchapp.ai/api/v1/tasks \
  -H "Authorization: Bearer $SWITCH_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "switch",
    "prompt": "a sunset over Miami",
    "count": 1
  }'

Authentication

Two kinds of credentials

Switch has two credential types. Which one you use depends on what you are calling, and full access uses both.

API key

sk_switch_live_…

Unlocks the REST API

Image generation and account info over /api/v1/*: the curl examples above, plus the CLI image and whoami commands.

Create one in the API keys section below, then send it as Authorization: Bearer.

OAuth sign-in

browser login

Unlocks the MCP tools

Video, uploads, library search, balance and model listing: the tools exposed over the MCP server.

Used by MCP clients and by switch login --oauth. You sign in through the browser; there is no key to copy.

Task creation

The request body

POST /api/v1/tasks takes a JSON body. Tasks are scoped to the key’s owner. Count caps at 10 per request.

Field	Type	Required	Default	Notes
model	string (alias)	yes	.	One of the public aliases. See Model aliases.
prompt	string	yes (or prompts)	.	Single prompt repeated count times.
prompts	string[]	yes (or prompt)	.	Explicit per-image prompts. Length sets count automatically.
count	integer	no	1	1 to 10. Ignored if prompts[] is set.
references	object[]	no	[]	{ url: "https://…", type: "face" \| "body" }
folder	string	no	"Generated"	Destination folder name in your Library.
quality	string	no	"2K"	0.5K \| 1K \| 2K \| 4K
scheduled_for	ISO 8601	no	null	The task waits and runs at this time.

API keys

Create and manage keys

Create a key and use it as your Bearer token. Each key is shown once, when you create it. Copy it then, because it can never be displayed again. You can keep up to 25 active keys.

Your keys comes with a paid membership.

See memberships

Revoke

Turn a key off

Click Revoke next to a key and confirm. It stops working immediately, and any further request with it returns:

HTTP/2 401
{ "error": { "code": "INVALID_KEY", "message": "Missing or invalid Switch API key. ..." } }

Revoking is permanent. Create a new key to replace it.

MCP connector

Let an AI assistant use Switch

Switch runs a hosted MCP server, so AI tools like Claude and Cursor can use Switch directly. There is nothing to install and no server to run. You point your client at the endpoint and sign in.

Endpoint

https://www.switchapp.ai/mcp

Connect

What it gives you

Once connected, the assistant can generate video, upload media, search your library, check your balance, and list available models, all against your own Switch account.

The MCP server is hosted by Switch, so you never run server code. Sign-in uses OAuth, or an API key from this page sent as a Bearer header for bots and scripts that cannot open a browser.

Voice & audio

Give your videos a voice

The voice library, `/api/audio/voices`

One library for every saved voice. Each voice comes back with a kind: reference is an instant voice match saved from a clip (free, unlimited); clone is a trained voice.

GET /api/audio/voicesList your voices, newest first. Query: limit (up to 200), offset, q (name/tag search), kind. Returns voices[] with voiceId, name, kind, ready, tags and a short-lived sampleUrl preview.
POST /api/audio/voicesSave a reference voice from a clip: name plus audioBase64 (with optional format) or a reachable audioUrl, or audioGenerationId to reuse a Library clip. Clip of about 5 to 30 seconds, max 10MB. Free, ready instantly.
PATCH /api/audio/voicesRename or re-tag: speakerId (the voiceId) plus name and/or tags.
DELETE /api/audio/voicesRemove a voice: speakerId. It disappears from every picker.

The routing rule

Speak, `POST /api/audio/tts`

Generative audio, `POST /api/audio/generate`

Examples

// Signed in at www.switchapp.ai — run in the browser console.
const r = await fetch('/api/audio/voices?limit=50', { credentials: 'include' });
const { voices } = await r.json();
// [{ voiceId: 'ref_…', name: 'My narrator', kind: 'reference',
//    ready: true, tags: [], sampleUrl: 'https://…' }, …]
// kind decides the speak route, see "The routing rule" above.
// Optional filters: ?q=narrator  ?kind=reference|clone  ?offset=50

From an AI assistant, the MCP tools

Connected through the MCP connector, four tools cover the whole flow:

voiceYour voice library in one tool, action: list, create (save a reference voice from audio_url or audio_base64 + voice_name), rename (voice_id takes the id or current name, plus new_name), clone (registers a talking-avatar voice, 2 credits), delete.
generate_audioSpeak any script. Pass the saved voice by NAME as voice, reference voices, trained voices and presets are all routed correctly by kind. Or match instantly from clips with reference_audio_url(s) (up to 3, @Audio1 to@Audio3 for dialogue). Knobs: speech_rate −50…100, pitch −12…12, loudness −50…100.
upload_mediaTurn a local or attached clip into a clean URL, pass url or base64 + mime, then feed the returned url into voice create or lip_sync_video as the source.
lip_sync_videoMake the person in a video speak your audio. Simplest path: action: "create" with engine: "best", video_url and sound_file (whole-clip sync). Precise path: identify-face → create with a face and timing → status. Source video mp4/mov, 2 to 60s, up to 100MB, 720p/1080p.

A lip sync render takes about 2 minutes; failed renders refund themselves. The finished video comes back at the source clip’s resolution and lands in your library and feed.

Rate limits

How much, how fast

Per user, image-counted. A count: 10 request uses 10 of your minute budget.

images / minute

1,000

images / day

max count / request

Reference

Command line

Run Switch from your terminal

The switch CLI runs Switch from your terminal, handy for scripts and automation. It calls the same APIs as everything else; there is no second account or backend.

Install

npm install -g @switchapp/cli

Developer preview. The package is not published to npm yet, so this command does not work today; it is how you will install the CLI once it ships.

Log in

# REST: image generation + whoami
switch login --key sk_switch_live_...

# MCP: video, uploads, library
switch login --oauth

See Authentication for which login unlocks which commands. Credentials are stored in ~/.switch/config.json.

Check your account

switch whoami

Generate an image

switch image generate --prompt "a sunset over Miami"
switch image status <task_id>

Troubleshooting

401 / invalid key: image generation needs an Ultra plan key. Create one in the API keys section below.
“unknown tool” on video, upload, or library: those run over MCP; sign in with switch login --oauth first.
402 / out of tokens: top up your account, then retry.

Add --json to any command for machine-readable output.

Usage & budgets

Read spend, set budgets

Read your spend and set budgets from code, the same data the Spend Center shows. Send your API key as Authorization: Bearer.

GET /api/v1/usage

Spend today / this week / this month, your effective limit, remaining, a forecast, and a breakdown by model and by source.

Plain English: how much you've spent, how much is left, and where it went.

GET /api/v1/budget

Your configured budget plus the effective caps actually in force (the most restrictive of your budget, your agency cap, and any admin limit).

Plain English: what your spending limits are right now.

POST /api/v1/budget

Plain English: set the most you want to spend per day, week, or month.

# From the CLI (developer preview)
switch usage              # today's spend, limit, remaining, forecast
switch budget set 200     # cap spending at $200/day
switch budget show        # see your current budget
switch budget disable     # remove your self-set budget

The CLI is in developer preview and not yet published to npm; these commands call the same endpoints above.

API examples

Curl examples

Replace $SWITCH_API_KEY with the key from your API keys section.

curl -X POST https://www.switchapp.ai/api/v1/tasks \
  -H "Authorization: Bearer $SWITCH_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "switch",
    "prompt": "a sunset over Miami",
    "count": 1
  }'

Models

Model aliases

Friendly aliases resolve to the live production model at request time. Pricing is never cached.

Alias	Model	Tokens / image
switch	Switch Model	1
switch_pro	Switch Pro	2
switch_ultra	Switch Ultra	5
nano_banana_2	Nano Banana 2	4
nano_banana_pro	Nano Banana Pro	6
gpt_image	GPT Image 2	3
gpt_image_2	GPT Image 2	3
gpt_image_transparent	GPT Image 2 (transparent)	3
gpt_image_25	GPT Image 2.5	1 / 3 / 8
gpt_image_25_flare	GPT Image 2.5 Flare	1 / 3 / 8

Errors

Error codes

Every error returns JSON shaped like { error: { code, message } }.

Status	Code	Meaning
400	INVALID_JSON	Body is not valid JSON.
400	MODEL_REQUIRED	Body missing the model field.
400	PROMPT_REQUIRED	Provide prompt (string) or prompts (string[]).
400	BAD_COUNT	count must be an integer between 1 and 10.
400	BAD_QUALITY	quality must be 0.5K, 1K, 2K, or 4K.
400	BAD_REFERENCE_TYPE	Reference type must be face or body.
400	BAD_REFERENCE_URL	Reference URL must be valid, public, and http(s).
400	UNKNOWN_ALIAS	Model alias is not in the public set.
400	INACTIVE_MODEL	Alias resolved but production model is no longer active.
400	BAD_SCHEDULE	scheduled_for must be ISO 8601.
401	INVALID_KEY	Missing, malformed, or revoked API key.
403	INSUFFICIENT_SCOPE	Key lacks the tasks:create scope.
403	ULTRA_REQUIRED	Account is not Switch Ultra (or admin).
429	RATE_LIMITED	Per-minute or per-day budget exhausted. Retry-After header included.
500	AUTH_INTEGRITY	Temporary server error. Retry; contact support if it persists.
500	ENQUEUE_FAILED	Could not queue the task. Please retry.
404	TASK_NOT_FOUND	Task status or receipt was not found for this key owner.

One API call. Real images.

Four ways in.

Generate from code

Connect AI assistants

Run it from the terminal

Bring your own look

Say it. Get it.

Know what you're spending

Your first call

Two kinds of credentials

API key

OAuth sign-in

The request body

Create and manage keys

Turn a key off

Let an AI assistant use Switch

Endpoint

Connect

What it gives you

Give your videos a voice

The voice library, /api/audio/voices

The routing rule

Speak, POST /api/audio/tts

Generative audio, POST /api/audio/generate

Examples

From an AI assistant, the MCP tools

How much, how fast

Run Switch from your terminal

Install

Log in

Check your account

Generate an image

Troubleshooting

Read spend, set budgets

Curl examples

Model aliases

Error codes

One API call. Real images.

Four ways in.

Generate from code

Connect AI assistants

Run it from the terminal

Bring your own look

Say it. Get it.

Know what you're spending

Your first call

Two kinds of credentials

API key

OAuth sign-in

The request body

Create and manage keys

Turn a key off

Let an AI assistant use Switch

Endpoint

Connect

What it gives you

Give your videos a voice

The voice library, /api/audio/voices

The routing rule

Speak, POST /api/audio/tts

Generative audio, POST /api/audio/generate

Examples

From an AI assistant, the MCP tools

How much, how fast

Run Switch from your terminal

Install

Log in

Check your account

Generate an image

Troubleshooting

Read spend, set budgets

Curl examples

Model aliases

Error codes

The voice library, `/api/audio/voices`

Speak, `POST /api/audio/tts`

Generative audio, `POST /api/audio/generate`

The voice library, `/api/audio/voices`

Speak, `POST /api/audio/tts`

Generative audio, `POST /api/audio/generate`