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: record H3 Max lip-sync audio ceiling verification against the queue schema (#11365)

[._.]/ Adam Eivybf3f7ef47d3c
diff --git a/server/lib/falVideoModels.js b/server/lib/falVideoModels.js
index 7beec72d51..34710d7249 100644
--- a/server/lib/falVideoModels.js
+++ b/server/lib/falVideoModels.js
@@ -26,7 +26,7 @@
*/
/** Date the schemas and prices below were last verified against fal.ai. */
-export const FAL_VIDEO_PRICING_VERIFIED_AT = '2026-10-11';
+export const FAL_VIDEO_PRICING_VERIFIED_AT = '2026-10-13';
/**
* What an entry renders from: text, a start frame, reference assets, a frame +
@@ -137,7 +137,10 @@ export const FAL_VIDEO_MODELS = Object.freeze([
// No prompt field at all: the frame and the audio are the whole request.
prompt: 'none',
image: { param: 'image_url', required: true },
- // fal accepts 5s to 15 minutes; the output length follows the audio.
+ // fal accepts 5s to 15 minutes; the output length follows the audio. The queue
+ // OpenAPI schema (checked 2026-10-13) says so; the h3-max-lip-sync marketing page
+ // claims a 15s clip and $0.255/s at 2K, which contradicts the schema and the model
+ // page. $0.32 stays: the higher figure is the safe side for the autopilot dollar cap.
audioInput: { param: 'audio_url', minSec: 5, maxSec: 900 },
duration: { kind: 'audio' },
resolution: { param: 'resolution', options: [...H3_RESOLUTIONS, '2K'], default: '768P' },