Silico Grove API Integration Guide

Integrate text, image, video, and audio capabilities through the endpoint matching each feature. Preserve request field names exactly.

Item	Value
Primary Base URL	https://ai.silicogrove.com/v1
Backup Base URL	https://api.silicogrove.com/v1 when the primary domain is unavailable
Authentication	Authorization: Bearer YOUR_API_KEY
JSON requests	Content-Type: application/json
File uploads	multipart/form-data

Start with Quick start, then confirm access in Models and access.

Quick Start

1. Create and securely store an API key in the console.
2. List the models available to that key.
3. Send requests to the matching capability endpoint.

Do not duplicate the Base URL

The OpenAI SDK base_url includes /v1. Do not append /v1 again when composing complete endpoint URLs. The backup address requires client-side failover configuration.

List available models

curl -X GET "https://ai.silicogrove.com/v1/models" \
  -H "Authorization: Bearer YOUR_API_KEY"

Minimal text request

curl -X POST "https://ai.silicogrove.com/v1/chat/completions" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.4-mini","messages":[{"role":"user","content":"Hello, reply in one sentence."}]}'

Python OpenAI SDK

from openai import OpenAI

client = OpenAI(api_key="YOUR_API_KEY", base_url="https://ai.silicogrove.com/v1")
response = client.chat.completions.create(
    model="gpt-5.4-mini",
    messages=[{"role": "user", "content": "Hello, reply in one sentence."}],
)
print(response.choices[0].message.content)

Models and Access

Available models depend on your account, group, and API key permissions. Treat GET /v1/models as the source of truth.

Capability	Typical models	Endpoint
Text	gpt-5.4-mini, Claude, Gemini	/v1/chat/completions
Images	gpt-image-2, Gemini image models, grok-imagine-image	Sync /v1/images/generations; async /v1/images/tasks
Video	Recommended: grok-video-1.5 (only 6, 8, 10, 12, or 15 seconds), kling-video-v3, video-ds-2.0, as-sd2.0-fast; grok-imagine-video, grok-imagine-video-1.5 (currently unavailable)	/v1/videos
Audio	Check the returned model list	/v1/audio/*

Text

Chat Completions

curl -X POST "https://ai.silicogrove.com/v1/chat/completions" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.4-mini","messages":[{"role":"system","content":"You are a concise assistant."},{"role":"user","content":"Write a product introduction."}],"stream":false}'

Responses API

curl -X POST "https://ai.silicogrove.com/v1/responses" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.4-mini","input":"Introduce yourself in one sentence."}'

Images

List models with your API key before use; its GET /v1/models response is the source of truth.

curl "https://ai.silicogrove.com/v1/models" \
  -H "Authorization: Bearer YOUR_API_KEY"

Model	Generate	Edit	OpenAI image API	Native Gemini API
gpt-image-2	Yes	Yes	Yes	No
gpt-image-2-all	Depends on visibility in /v1/models	Depends on upstream	Yes	No
gemini-3-pro-image	Yes	Yes	Yes	Yes
gemini-3.1-flash-image	Yes	Yes	Yes	Yes
gemini-3-pro-image-preview	Yes	Yes	Yes	Yes
gemini-3.1-flash-image-preview	Yes	Yes	Yes	Yes
gemini-2.5-flash-image	Yes	Yes	Yes	Yes
grok-imagine-image, grok-imagine-image-pro	Depends on visibility in /v1/models	Depends on upstream	Yes	No

Use https://ai.silicogrove.com as the primary Base URL and https://api.silicogrove.com as an explicit backup. All requests use Authorization: Bearer YOUR_API_KEY.

OpenAI-compatible generation

This interface works with gpt-image-2 and Gemini image models.

curl "https://ai.silicogrove.com/v1/images/generations" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2",
    "prompt": "A premium product poster, realistic photography, clean background",
    "size": "1024x1024",
    "quality": "high",
    "n": 1,
    "response_format": "url"
  }'

Gemini model example:

curl "https://ai.silicogrove.com/v1/images/generations" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-3.1-flash-image",
    "prompt": "A 16:9 neon city in the rain at night",
    "size": "1792x1024",
    "quality": "2k",
    "n": 1,
    "response_format": "url"
  }'

OpenAI-compatible editing

Use multipart form data. Do not set Content-Type manually; curl or your SDK must generate the multipart boundary.

curl "https://ai.silicogrove.com/v1/images/edits" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "model=gpt-image-2" \
  -F "prompt=Keep the subject and change the background to a city at night" \
  -F "image=@/path/to/input.png" \
  -F "size=1024x1024" \
  -F "quality=high" \
  -F "n=1"

For Gemini, change model to gemini-3.1-flash-image and use quality=2k.

Supported input formats are PNG, JPG, JPEG, and WebP. The playground currently accepts one reference image per request, up to 10 MB.

OpenAI image parameters

The gateway accepts the common fields below. Individual models may support only a subset; an upstream may ignore or reject unsupported values. Treat /v1/models and the model's actual response as authoritative.

Field	Type	Required	Common values or limits	Description
model	string	Yes	From GET /v1/models	Image model name.
prompt	string	Yes	Natural-language text	Generation or editing instruction.
n	integer	No	1 to 10; default 1	Number of images. Gemini image models require 1; other models may impose lower limits.
size	string	No	Common: 1024x1024, 1536x1024, 1024x1536, 1792x1024, 1024x1792	Size or aspect-ratio mapping is model-dependent. DALL-E models enforce narrower enums.
quality	string	No	Common: auto, low, medium, high, standard, hd, 1k, 2k, 4k	Quality or output-resolution tier. Upstream case sensitivity may differ.
response_format	string	No	url, b64_json	Desired synchronous response. Async tasks always return a persisted URL.
background	string	No	auto, opaque, transparent	Background mode; model-dependent.
output_format	string	No	png, jpeg, webp	Output file format; model-dependent.
output_compression	integer	No	Usually 0 to 100	JPEG/WebP compression quality; model-dependent.
style	string	No	Common: vivid, natural	Style option; model-dependent.
moderation	string	No	Common: auto, low	Moderation level; model-dependent.
stream	boolean	No	true, false	Requests a streamed synchronous response. Async tasks require false or omission.
image, image[]	file or model-supported JSON URL/value	For editing	PNG, JPEG, WebP; multipart file limit 10 MB	Reference image. Async multipart files are persisted before worker execution.
mask	file or model-supported JSON URL/value	No	Usually PNG	Edit mask; model-dependent.
input_fidelity	string	No	Common: low, high	Reference-image fidelity; model-dependent.

With curl -F, do not set Content-Type manually. Send scalar multipart values such as n and output_compression as text form fields.

OpenAI-compatible response

When R2 upload succeeds, a URL is returned:

{"created":1786192453,"data":[{"url":"https://file.lunadownload.com/temporary/2026/08/12/uuid.png"}]}

If R2 is not configured or upload fails, the response contains b64_json instead:

{"created":1786192453,"data":[{"b64_json":"iVBORw0KGgo..."}]}

Field	Type	Description
created	integer	Image creation timestamp returned by the provider.
data	array	Image result list.
data[].url	string	R2 or provider image URL. The URL may be periodically removed.
data[].b64_json	string	Base64 image, normally present only for synchronous requests when R2 rewriting is unavailable.
data[].revised_prompt	string	Provider-revised prompt; returned only by some models.

Temporary image URLs

Images returned through file.lunadownload.com are temporary and periodically removed. Download and store generated files promptly.

Asynchronous image tasks

Use the asynchronous API when generation may exceed client or reverse-proxy timeouts. Submission immediately returns a task ID; a background worker calls the provider and persists the result to R2. Existing synchronous endpoints remain unchanged.

Operation	Endpoint	Request format
Async generation	POST /v1/images/tasks	JSON without image, images, or mask
Async editing	POST /v1/images/tasks	Multipart file upload, or JSON containing image fields
Retrieve task	GET /v1/images/tasks/{task_id}	GET

R2 storage must be enabled for async tasks, and stream: true is unsupported. Multipart reference images are limited to 10 MB each. A task becomes completed only after its output is persisted to R2. Provider or storage failures mark it failed and refund the pre-consumed quota.

Async generation

curl -X POST "https://ai.silicogrove.com/v1/images/tasks" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2",
    "prompt": "A premium product poster in a realistic photography style",
    "size": "1024x1024",
    "quality": "high",
    "n": 1
  }'

Async editing

curl -X POST "https://ai.silicogrove.com/v1/images/tasks" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "model=gpt-image-2" \
  -F "prompt=Keep the subject and change the background to a city at night" \
  -F "image=@/path/to/input.png" \
  -F "size=1024x1024" \
  -F "quality=high" \
  -F "n=1"

Submission response

A successful submission returns HTTP 202 Accepted:

{
  "id": "task_0123456789abcdef",
  "object": "image.task",
  "status": "queued",
  "progress": "0%",
  "created_at": 1786192453,
  "started_at": 0,
  "completed_at": 0
}

Retrieve and poll

curl "https://ai.silicogrove.com/v1/images/tasks/task_0123456789abcdef" \
  -H "Authorization: Bearer YOUR_API_KEY"

Only the user who created a task can retrieve it. Poll every 2 to 5 seconds; avoid high-frequency concurrent polling.

Field	Type	Present when	Description
id	string	Always	Public task ID in task_... format.
object	string	Always	Always image.task.
status	string	Always	queued, processing, completed, or failed.
progress	string	Always	Percentage string such as 0%, 10%, or 100%.
created_at	integer	Always	Task submission Unix timestamp.
started_at	integer	Always	Execution start Unix timestamp, or 0 before execution.
completed_at	integer	Always	Terminal Unix timestamp, or 0 before completion.
created	integer	Success	Creation timestamp from the image result.
data	array	Success	Image results; read the output from data[].url.
error.message	string	Failure	Task failure reason.
error.type	string	Failure	Currently image_task_failed.

status	Meaning	Terminal
queued	Waiting for a worker	No
processing	Calling the provider or storing output	No
completed	Finished; data[].url is available	Yes
failed	Failed; inspect error	Yes

Completed response:

{
  "id": "task_0123456789abcdef",
  "object": "image.task",
  "status": "completed",
  "progress": "100%",
  "created_at": 1786192453,
  "started_at": 1786192455,
  "completed_at": 1786192550,
  "created": 1786192548,
  "data": [{"url": "https://file.lunadownload.com/temporary/2026/08/12/uuid.png"}]
}

Failed response:

{
  "id": "task_0123456789abcdef",
  "object": "image.task",
  "status": "failed",
  "progress": "100%",
  "created_at": 1786192453,
  "started_at": 1786192455,
  "completed_at": 1786192550,
  "error": {"message": "upstream request failed", "type": "image_task_failed"}
}

Async HTTP status codes

Status	Scenario
202	Task created.
400	Invalid parameters, multipart file, n, or stream.
403	API key, quota, group, or model access denied.
404	Task is absent, has a different type, or belongs to another user.
429	Request or model concurrency limit reached.
503	R2 storage required by async tasks is not configured or available.

Submission or retrieval failures that occur before a task response use this error envelope:

{
  "error": {
    "message": "R2 storage is required for asynchronous image tasks",
    "type": "storage_unavailable",
    "code": "storage_unavailable"
  }
}

There is currently no /v1/images/tasks/{task_id}/content endpoint. Download data[].url from the completed response directly.

Native Gemini generation

This interface is only for Gemini image models. gpt-image-2 does not support it.

curl "https://ai.silicogrove.com/v1beta/models/gemini-3.1-flash-image:generateContent" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{"role":"user","parts":[{"text":"Generate a 16:9 neon city in the rain at night"}]}],
    "generationConfig": {
      "responseModalities": ["TEXT", "IMAGE"],
      "imageConfig": {"aspectRatio":"16:9","imageSize":"2K"}
    }
  }'

Native Gemini editing

Gemini has no separate native editing endpoint. Send the reference image as inlineData to the same generateContent endpoint.

IMAGE_B64=$(base64 < input.png | tr -d '\n')

curl "https://ai.silicogrove.com/v1beta/models/gemini-3.1-flash-image:generateContent" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{
    \"contents\":[{\"role\":\"user\",\"parts\":[
      {\"text\":\"Keep the subject and change the background to a city at night\"},
      {\"inlineData\":{\"mimeType\":\"image/png\",\"data\":\"${IMAGE_B64}\"}}
    ]}],
    \"generationConfig\":{\"responseModalities\":[\"TEXT\",\"IMAGE\"],\"imageConfig\":{\"aspectRatio\":\"1:1\",\"imageSize\":\"2K\"}}
  }"

When R2 upload succeeds, generated media appears as fileData.fileUri; otherwise Gemini returns the original inlineData base64 payload.

Parameter mapping for Gemini models

OpenAI image parameter	Native Gemini field
prompt	contents[].parts[].text
image file	contents[].parts[].inlineData
size: 1024x1024	aspectRatio: 1:1
size: 1792x1024	aspectRatio: 16:9
size: 1024x1792	aspectRatio: 9:16
size: 1536x1024	aspectRatio: 3:2
size: 1024x1536	aspectRatio: 2:3
quality: auto, fast, or 1k	imageSize: 1K
quality: high, hd, or 2k	imageSize: 2K
quality: 4k	imageSize: 4K

Gemini image models currently generate one image per request: set n to 1.

Recommended integration

Use the OpenAI-compatible endpoints for standard clients and relay integrations:

POST /v1/images/generations
POST /v1/images/edits
POST /v1/images/tasks
GET /v1/images/tasks/{task_id}

Use POST /v1beta/models/{model}:generateContent only when you need the complete Gemini request format, multimodal contents, or the Gemini SDK.

Video

Video generation is asynchronous. Submit to POST /v1/videos, retain the returned task ID, then poll GET /v1/videos/{task_id} until the task reaches a terminal state. Video models must not be sent to /v1/chat/completions.

Available models depend on the API key and group. Call GET /v1/models before presenting model choices to end users.

Recommended model: Grok Video 1.5

Model	Duration	Resolution
grok-video-1.5	15 seconds by default; 6, 8, 10, 12, or 15 seconds when specified	Optional; 720p recommended

grok-video-1.5 supports text-to-video, a single reference image, and multiple reference images (up to 7). The three examples below use the same endpoint; add image_urls when reference images are needed.

WARNING

When seconds is omitted, the request generates a 15-second video by default. When specified, seconds must be the string "6", "8", "10", "12", or "15". Sending "4" or any other value returns: seconds must be one of: 6, 8, 10, 12, 15.

Other supported Kling V3 models

Model	Duration	Resolution	Billing
kling-video-v3	3-15 seconds	720p, 1080p, 4k	Per second and selected resolution
kling-video-v3-omni	3-15 seconds	720p, 1080p, 4k	Per second and selected resolution
kling-video-v3-turbo	3-15 seconds	720p, 1080p	Per second and selected resolution

Send resolution as a top-level field in Kling create requests. Do not use the image-generation quality field for video resolution. The effective price is determined when the task is submitted, so do not rely on a fixed documented amount.

Grok Video 1.5 examples

Every example uses POST /v1/videos, Authorization: Bearer YOUR_API_KEY, and Content-Type: application/json.

1. Text to video

curl -X POST "https://ai.silicogrove.com/v1/videos" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "grok-video-1.5",
    "prompt": "A cinematic sunrise over a futuristic coastal city, slow aerial camera movement",
    "seconds": "15",
    "aspect_ratio": "16:9",
    "resolution": "720p"
  }'

2. One reference image

{
  "model": "grok-video-1.5",
  "prompt": "Rotate the product smoothly in soft studio lighting, as a clean premium commercial",
  "seconds": "10",
  "aspect_ratio": "9:16",
  "resolution": "720p",
  "image_urls": ["https://example.com/product.png"]
}

3. Multiple reference images

{
  "model": "grok-video-1.5",
  "prompt": "Create a coherent product showcase using these references, with cinematic camera movement and premium commercial lighting",
  "seconds": "15",
  "aspect_ratio": "16:9",
  "resolution": "720p",
  "image_urls": [
    "https://example.com/product-front.png",
    "https://example.com/product-detail.png"
  ]
}

The response includes a public id or task_id. Store that value; do not use an upstream task ID obtained from another API response.

{
  "id": "task_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx",
  "object": "video",
  "status": "queued"
}

This example uses the recommended grok-video-1.5 model to create a 15-second text-to-video task. A successful create request returns an asynchronous task. Store task_id, then use Poll a task to retrieve progress and output.

{
  "id": "task_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx",
  "task_id": "task_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx",
  "object": "video.generation",
  "model": "grok-video-1.5",
  "status": "queued",
  "progress": 0,
  "created_at": 1786869597,
  "result": {}
}

Reference assets

Use public HTTPS URLs or complete data: URLs such as data:image/png;base64,.... Do not send a local file path or bare base64 bytes. Local files can be uploaded through Reference assets first.

image_urls is the preferred field. images is accepted for compatibility. Send only one of them. reference_images and input_reference: {"image_url":"..."} are also accepted for integrations that use those names; do not combine reference_images and input_reference in one request.

Request fields

Field	Required	Description
model	Yes	A video model available to the API key.
prompt	Yes	Describe the subject, motion, camera, visual style, and composition.
seconds	No	Requested duration as a string. When omitted, grok-video-1.5 defaults to 15 seconds. When specified, it accepts only "6", "8", "10", "12", or "15".
aspect_ratio	No	16:9, 9:16, or 1:1.
resolution	Required for Kling V3; optional for Grok	720p is recommended for Grok. Kling supports 480p, 720p, 1080p, or 4k, subject to the selected model. Send it as a top-level field; do not use quality as a replacement.
image_urls	No	Preferred array of up to 7 reference image URLs or complete data URLs.
images	No	Compatibility alias for image_urls; do not send both.
image	No	grok-imagine-video-1.5 first-frame mode only. A single image URL or complete data URL. Do not combine with reference_images.
reference_images	No	Compatibility array of reference images; do not combine with input_reference.
input_reference	No	Compatibility single-image form: { "image_url": "https://..." }.

The older video-ds-* models support images, videos, and audios. Their media limits are 4 images, 3 videos, and 1 audio file.

Poll a task

curl -X GET "https://ai.silicogrove.com/v1/videos/TASK_ID" \
  -H "Authorization: Bearer YOUR_API_KEY"

Treat queued and in_progress as non-terminal. completed means the video is available; failed means generation ended unsuccessfully. Poll no more frequently than once every 5 seconds, and retain the task ID if a client-side timeout is reached.

Download the completed video

curl -L "https://ai.silicogrove.com/v1/videos/TASK_ID/content" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  --output result.mp4

The service may return a signed result URL internally. It is temporary and should be downloaded promptly. Use the content endpoint above rather than reconstructing an upstream URL, task ID, domain, or signature.

Security and reliability recommendations

API keys and callers

• Store API keys only in server-side environment variables or a secrets manager. Browsers, mobile apps, and public frontend code must not hold long-lived keys.
• Use separate API keys per application or purpose. A compromised key can then be revoked without interrupting unrelated workloads.
• After creating a task, persist both your local business identifier and the returned task_id for polling, reconciliation, and retry control.

Reference assets and downloads

• Reference URLs must be safely reachable by the video service. Do not provide private-network IP addresses, administrative endpoints, cloud credential URLs, or local file paths.
• After completion, download videos through the authenticated /content endpoint rather than exposing temporary result URLs long term.
• Create a replacement task only after the existing task explicitly returns failed. Continue polling the same task_id while it is queued or in_progress to prevent duplicate generations and charges.

Grok Imagine models (currently unavailable)

WARNING

grok-imagine-video and grok-imagine-video-1.5 are currently unavailable and must not be used for production requests. This section is retained as a historical parameter reference; use the active Kling models above.

Model	Generation modes	Duration	Resolution
grok-imagine-video	Text to video	Model-dependent	Model-dependent
grok-imagine-video-1.5	Text to video, first-frame image to video, reference-image video	4, 6, 8, 10, 12, or 15 seconds	Text and first-frame: 480p, 720p, 1080p; reference images: up to 720p

grok-imagine-video-1.5 has two mutually exclusive image modes: first-frame mode accepts one image URL; reference-image mode accepts 1-7 reference_images URLs and uses <IMAGE_1>, <IMAGE_2>, and similar placeholders in the prompt. Do not send image, images, image_urls, or input_reference with reference_images. Reference-image mode is limited to 720p.

Historical request example:

{
  "model": "grok-imagine-video-1.5",
  "prompt": "The person from <IMAGE_1> walks through a city street in a cinematic commercial shot",
  "reference_images": ["https://example.com/person.jpg"],
  "seconds": "8",
  "aspect_ratio": "16:9",
  "resolution": "720p"
}

Reference Assets

Upload local images, videos, and audio to the temporary asset endpoint. The returned data.url can be placed in a video request's images, videos, or audios array.

Temporary files are periodically removed

Uploaded assets and generated results are temporary. Their URLs may expire during periodic cleanup. Download and store required files promptly; do not treat returned URLs as permanent storage.

# Image
curl -X POST "https://ai.silicogrove.com/pg/assets" -H "Authorization: Bearer YOUR_API_KEY" -F "kind=image" -F "file=@/path/to/ref.jpg"

# Video
curl -X POST "https://ai.silicogrove.com/pg/assets" -H "Authorization: Bearer YOUR_API_KEY" -F "kind=video" -F "file=@/path/to/ref.mp4"

# Audio
curl -X POST "https://ai.silicogrove.com/pg/assets" -H "Authorization: Bearer YOUR_API_KEY" -F "kind=audio" -F "file=@/path/to/ref.mp3"

Response example

{"success":true,"data":{"kind":"image","url":"https://file.lunadownload.com/temporary/2026/08/11/xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx.jpg","filename":"ref.jpg","content_type":"image/jpeg","size":123456}}

kind	Formats	Maximum file size
image	jpg, png, webp	10 MiB
video	mp4, mov, webm	100 MiB
audio	mp3, m4a, wav, aac, ogg, webm	20 MiB

Audio

Speech synthesis

curl -X POST "https://ai.silicogrove.com/v1/audio/speech" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"YOUR_TTS_MODEL","input":"Welcome to Silico Grove API.","voice":"alloy","response_format":"mp3"}' \
  --output speech.mp3

Transcription and translation

curl -X POST "https://ai.silicogrove.com/v1/audio/transcriptions" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "model=YOUR_STT_MODEL" \
  -F "file=@/path/to/audio.mp3" \
  -F "response_format=json"

curl -X POST "https://ai.silicogrove.com/v1/audio/translations" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "model=YOUR_STT_MODEL" \
  -F "file=@/path/to/audio.mp3" \
  -F "response_format=json"

Transcription and translation are multipart requests. The file field is file.

Troubleshooting

Symptom	Likely cause	Resolution
Model unavailable	The model and group do not match, or the key lacks permission.	Call /v1/models with the same key to confirm the model name.
model is required	A relay renamed the field.	Preserve model; do not change it to model_name.
Video fails or enters a chat model	The request was sent to a chat endpoint.	Call POST /v1/videos.
Reference asset has no effect	A local path was sent, or the value was not an array.	Send public URLs in images, videos, or audios.
Invalid seconds type	The duration was sent as a number.	Use a string, such as "seconds": "15".
Upload fails	Wrong content type or file field name.	Use -F and name the file field file.

When reporting an issue, include request time, model, endpoint, group, request ID, and the error response. Do not share API keys, authorization headers, or sensitive prompts.

Relay Integrations

• Use https://ai.silicogrove.com or https://ai.silicogrove.com/v1 as the primary upstream, depending on whether your relay adds /v1 automatically.
• The backup upstream is https://api.silicogrove.com or https://api.silicogrove.com/v1; configure failover in the relay.
• Synchronize /v1/models with a Silico Grove API key rather than entering unavailable models manually.
• Video models must stay on /v1/videos; do not add them to a chat model pool.
• Preserve the images, videos, and audios arrays when forwarding video requests.
• Image editing and audio transcription are multipart requests; do not lose their file fields during forwarding.
• Submit asynchronous images to POST /v1/images/tasks and poll GET /v1/images/tasks/{task_id}; do not wrap an asynchronous task as a synchronous image response.