gemini-interactions/gemini-omni-flash-preview, served by Comfy Router from Gemini Interactions.
Quick start
Create a key at platform.comfy.org/profile/api-keys and export it asCOMFY_API_KEY. The Python and TypeScript snippets use the Comfy SDKs (pip install comfy-sdk, npm install @comfyorg/sdk); the cURL snippet is the same call over raw HTTP.
Model ID: gemini-interactions/gemini-omni-flash-preview
Endpoint: POST https://api.comfy.org/v2/models/gemini-interactions/gemini-omni-flash-preview
Schema
Input
Router has not published an authored input schema for this model yet:
GET /v2/models/gemini-interactions/gemini-omni-flash-preview/openapi.json returns an open object with x-comfy-input-schema-authored: false. Router forwards the body to Gemini Interactions unchanged, so Gemini Interactions’s own API reference is authoritative for the request fields, and nothing is validated server side.Output
string
string
string
string
required
Always
completed on a Router response. The provider’s other statuses (in_progress, requires_action, failed, cancelled, incomplete, budget_exceeded) do not reach a caller through Router as a 200; they are returned as a Comfy Router error carrying the provider’s body.Possible values: completedobject[]
required
The interaction timeline, in order. Narrowed here from the untyped
steps on GeminiInteraction so the output leaf is addressable; the provider keeps adding step types, so an item is additionalProperties: true and only the fields a Router caller reads are declared.object[]
The typed content blocks of this step.
string
Base64-encoded inline media, on a media block delivered inline. Google caps inline media at 4 MB and requires
delivery: uri above it.string
Media type of
data or uri, on a media block.string
The generated text. Present on a
text block, and this is the leaf the nightly Router SDK case asserts (steps[].content[].text).string
Block kind —
text, image, audio, video or document.string
Reference to media delivered out of band, fetched by the caller from the URI it names.
string
Step kind.
model_output is the generated answer; user_input is the caller’s own turn echoed back by the retrieval route; thought is internal reasoning. The tool steps (function_call, function_result, code_execution_call, google_search_call, …) are open-ended and Google adds to them.object
Token usage for a Gemini interaction.
object[]
Token count for one modality.
string
One of
text, image, audio, video, document.integer
object[]
Token count for one modality.
string
One of
text, image, audio, video, document.integer
integer
integer
integer
integer
integer
Examples
Output
Before you ship
The snippets above are the shortest working call. Three things are the same for every model and are documented once on the Comfy Router headers page: send anIdempotency-Key on every paid call and reuse it when you retry, expect the connection to be held up to Router’s 10 minute deadline, and keep X-Comfy-Request-Id from every response. The SDKs do all three for you; the cURL tab does none of them. On failure, X-Comfy-Error-Type names the bucket, and a 422 means the body failed the model’s schema and was never billed.
Headers
Authentication, idempotency, request IDs, error buckets, retry pacing, spend limits.
Quick Start
Typed error handling in Python and TypeScript, reading the 422, walking the catalog.
Limitations
What Router does not do today, and what to use instead.