Skip to content

Self-hosted, Self-Improving, Federated Everything App — dev management, agent orchestration, persistent mind, digital twin, second brain, and creative pipeline from a single dashboard over Tailscale to port the user into the OS

docs: describe Runway metered image routing (#13105)

Adam Eivya538cd145a37
diff --git a/server/lib/generationModes.js b/server/lib/generationModes.js
index 2697abb7ec..0918820cb5 100644
--- a/server/lib/generationModes.js
+++ b/server/lib/generationModes.js
@@ -20,7 +20,7 @@ export const IMAGE_GEN_MODE = Object.freeze({
export const IMAGE_GEN_MODES = Object.freeze(Object.values(IMAGE_GEN_MODE));
-// Cloud image backends spend remote quota (or money, for fal.ai) and run
+// Cloud image backends spend remote quota or metered credit and run
// through the media queue's parallel cloud lane rather than the local
// accelerator lane.
export const CLOUD_IMAGE_GEN_MODES = Object.freeze([
diff --git a/server/services/imageGen/modes.js b/server/services/imageGen/modes.js
index d782a12dcf..db0d999d3f 100644
--- a/server/services/imageGen/modes.js
+++ b/server/services/imageGen/modes.js
@@ -177,7 +177,7 @@ export const nearestAgyAspectRatio = (width, height) => nearestAspectRatio(width
* Resolve the queue-capable image mode for a render request: the per-request
* override (honored only when that backend is enabled/available), else the
* saved dispatcher default, else the first enabled cloud backend in
- * CLOUD_IMAGE_GEN_MODES order (codex → grok → agy → fal) → local. External
+ * CLOUD_IMAGE_GEN_MODES order (codex → grok → agy → fal → runway) → local. External
* never queues. Hoisted from the pipeline visual stages (#2896) so sprite renders and
* any future queued surface share one enable-gating ladder — see issue #2881
* for the wider param-assembly consolidation.
diff --git a/server/services/imageGenQuota.js b/server/services/imageGenQuota.js
index 2358efb43e..7c6a830793 100644
--- a/server/services/imageGenQuota.js
+++ b/server/services/imageGenQuota.js
@@ -53,14 +53,13 @@ const UNKNOWN_BLOCK_TTL_MS = 60 * 60 * 1000;
// Only the cloud CLIs spend remote image quota — local renders run on the
// user's own GPU and external hits their own SD endpoint, so neither has
// anything to report. Keyed off CLOUD_IMAGE_GEN_MODES rather than a second
-// hand-maintained list, so a 4th cloud backend is tracked the moment it is
+// hand-maintained list, so a subscription backend is tracked the moment it is
// added there. Display names come from `modeLabel` for the same reason — one
// backend must not appear under two names across the UI.
//
-// fal.ai is the one cloud backend excluded: it is billed per image on a metered
-// account, not drawn from a subscription quota window, so there is no reset to
-// observe and its failures (content policy, bad input) are not quota blocks.
-// Its spend is recorded per render instead (`estimatedCostUsd` in the sidecar).
+// Metered fal.ai and Runway images are excluded: they do not draw from a
+// subscription quota window. Policy/input failures must not become quota blocks.
+// Their spend is recorded per render (`estimatedCostUsd` in the sidecar).
export const isQuotaTrackedImageMode = (mode) => CLOUD_IMAGE_GEN_MODES.includes(mode) && ![IMAGE_GEN_MODE.FAL, IMAGE_GEN_MODE.RUNWAY].includes(mode);
/** Card id for the image-gen quota card, in the same namespace as the provider