Gemini Developer API provider contract#

Hubu ships two deliberately narrow Gemini Developer API contracts beside the FLUX contract. Operators select either or both contracts and provide only an opaque Keychain credential reference; the renderer supplies each immutable target, transport, capability, policy, and price.

Start with live provider operations for the shared credential, governance, pricing, spend, retry, reconciliation, artifact, and qualification procedure. This page is the canonical description of Gemini provider behavior; field definitions live in the providers.toml reference.

Frozen field	Value
Contracts	`hubu.gemini-3.1-flash-lite-image.text-to-image/v1`; `hubu.gemini-3.1-flash-image.text-to-image/v1`
Targets	`google` / `gemini_developer_image` / `gemini-3.1-flash-lite-image`; `google` / `gemini_developer_image` / `gemini-3.1-flash-image`
API	`https://generativelanguage.googleapis.com/` / `v1beta`
Lite capability	one 1024×1024 (`1k`) PNG or JPEG image
Non-Lite capability	one 1024×1024 (`1k`), 2048×2048 (`2k`), or 4096×4096 (`4k`) PNG or JPEG image
Price review	2026-09-01
Lite standard price	USD $0.0336 (`336/100` cents) at 1K
Non-Lite standard prices	USD $0.067 (`67/10` cents) at 1K; $0.101 (`101/10` cents) at 2K; $0.151 (`151/10` cents) at 4K
Retry and fallback	zero generation retries; no fallback
Transport	synchronous inline response; no polling

The stable models, resolution limits, and standard output prices come from Google's Lite model page, non-Lite model page, image generation guide, and Developer API pricing. Changing any frozen fact requires a new immutable contract and pricing version.

Configure#

Use an explicit composite catalog version when multiple contracts are enabled. Both Gemini contracts may use the same Google credential alias; FLUX must use an independently isolated credential:

schema_version = 1
mode = "live"
catalog_version = "operator-gemini-flux-2026-09-01-v1"
maximum_spend_minor = 25
live_spend_acknowledgement = "I_ACKNOWLEDGE_LIVE_PROVIDER_SPEND"

[[contract_bindings]]
contract = "hubu.gemini-3.1-flash-lite-image.text-to-image/v1"
credential = "google_gemini"

[[contract_bindings]]
contract = "hubu.gemini-3.1-flash-image.text-to-image/v1"
credential = "google_gemini"

[[contract_bindings]]
contract = "hubu.flux-2-pro.text-to-image/v1"
credential = "bfl_flux"

Credential coordinates must remain independent across providers. Readiness checks do not call a provider; see readiness and live qualification.

Execute and recover#

Discover the target with GET /v2/execution-targets or gongbu_list_execution_targets, then submit only its opaque target_id plus a normalized request. Use the Lite target only for 1k; use the non-Lite target for 1k, 2k, or 4k. The adapter verifies the selected size against the frozen schema-v2 pricing selector before calling the provider and never derives the authorized price from returned artifact dimensions.

The adapter reads an AI Studio API key from Keychain and sends it only in the x-goog-api-key header to the configured Google API endpoint. The default ProviderAdapter lifecycle performs one synchronous submission. A successful response completes immediately; an ambiguous timeout has no pollable checkpoint and requires reconciliation, not automatic resubmission.

FLUX uses the same authorization, pricing, claim, settlement, replay, artifact, and redaction lifecycle but overrides submission and polling. See the FLUX provider contract for that asynchronous transport.

Ordinary tests and demos remain fixture-only. Do not perform paid qualification without a separate, explicit operator-authorized procedure; the focused live test is documented in live provider operations.