@hraness/slopcamera 3.3.2 → 3.3.4

package/apps/desktop/cli/gateway-media-catalog.ts

@@ -920,3 +920,60 @@ export function createFileGatewayMediaCatalogSnapshotStore(
     },
   };
 }
+
+/**
+ * Alibaba Wan video models advertise normalized catalog resolutions such as
+ * "720p", but the provider API accepts only explicit pixel sizes bound to
+ * the requested aspect ratio ("1280x720"). This adapter translates the
+ * catalog value through Wan's documented size table so local qualification
+ * and duration pricing stay anchored to the catalog row while paid dispatch
+ * sends the provider representation. The `1280x720` (16:9) cell and the
+ * `480x832` enumeration hint are verified against the live provider; the
+ * remaining cells mirror Wan's published size enum.
+ */
+const WAN_RESOLUTION_PIXEL_SIZES: Readonly<
+  Record<string, Readonly<Record<string, string>>>
+> = {
+  "1080p": {
+    "16:9": "1920x1080",
+    "9:16": "1080x1920",
+    "1:1": "1440x1440",
+    "3:4": "1248x1632",
+    "4:3": "1632x1248",
+  },
+  "480p": {
+    "16:9": "832x480",
+    "9:16": "480x832",
+    "1:1": "624x624",
+  },
+  "720p": {
+    "16:9": "1280x720",
+    "9:16": "720x1280",
+    "1:1": "960x960",
+    "3:4": "832x1088",
+    "4:3": "1088x832",
+  },
+};
+
+/**
+ * Resolve the resolution value a provider expects on the wire.
+ * Returns `undefined` when the request carries no resolution, `null` when a
+ * Wan catalog resolution cannot be mapped for the requested aspect ratio,
+ * and the wire-ready value otherwise.
+ */
+export function providerVideoResolution(
+  modelId: string,
+  resolution: string | undefined,
+  aspectRatio: string | undefined,
+): string | null | undefined {
+  if (
+    resolution === undefined
+    || !modelId.startsWith("alibaba/wan-")
+    || !/^[1-9]\d{2,4}p$/iu.test(resolution)
+  ) {
+    return resolution;
+  }
+  // Wan produces 16:9 landscape output when a request omits the aspect ratio.
+  return WAN_RESOLUTION_PIXEL_SIZES[resolution.toLocaleLowerCase("en-US")]
+    ?.[aspectRatio ?? "16:9"] ?? null;
+}

package/apps/desktop/cli/gateway-media-service.ts

@@ -16,6 +16,7 @@ import {
 } from "./gateway-media-artifacts";
 import {
   GATEWAY_MEDIA_CATALOG_URL,
+  providerVideoResolution,
   type GatewayJsonValue,
   type GatewayMediaCatalogView,
   type GatewayMediaKind,
@@ -2930,6 +2931,12 @@ export function createGatewayMediaService(options: Readonly<{
         execution.signal,
       );
       assertVideoCapabilities(catalog.model, request);
+      const wireResolution = providerVideoResolution(
+        catalog.model.id,
+        request.resolution,
+        request.aspectRatio,
+      );
+      if (wireResolution === null) unsupportedModelSetting();
       const dispatchAt = now();
       const credential = await options.loadCredential(dispatchAt);
       const timeout = combineAbortSignals(
@@ -2989,9 +2996,9 @@ export function createGatewayMediaService(options: Readonly<{
             ...(request.providerOptions === undefined
               ? {}
               : { providerOptions: request.providerOptions }),
-            ...(request.resolution === undefined
+            ...(wireResolution === undefined
               ? {}
-              : { resolution: request.resolution }),
+              : { resolution: wireResolution }),
             ...(request.seed === undefined ? {} : { seed: request.seed }),
           });
           return parseGeneratedFiles(sdkResult, "videos", ["video/"]);
@@ -3025,6 +3032,9 @@ export function createGatewayMediaService(options: Readonly<{
             generateAudio: optionalSummaryValue(request.generateAudio),
             maxVideosPerCall: optionalSummaryValue(request.maxVideosPerCall),
             n: optionalSummaryValue(request.n),
+            providerResolution: optionalSummaryValue(
+              wireResolution === request.resolution ? undefined : wireResolution,
+            ),
             resolution: optionalSummaryValue(request.resolution),
             seed: optionalSummaryValue(request.seed),
           }), request.prompt, request.providerOptions),

package/apps/desktop/cli/help.ts

@@ -91,7 +91,7 @@ For local SDK workflows, code run, workflows run and runs resume accept
 The fixed local operation is slopcamera.studio.run; its SDK surface is studio.run().`,
   direct: `Usage:
   slopcamera direct init <recipe.json> [--json]
-  slopcamera direct anchor --input <image> [--json]
+  slopcamera direct anchor --input <media> [--json]
   slopcamera direct plan <recipe.json> [--json]
   slopcamera direct start <recipe.json> --budget-usd <USD> [--json]
   slopcamera direct inspect <direct-id> [--json]
@@ -105,13 +105,13 @@ The fixed local operation is slopcamera.studio.run; its SDK surface is studio.ru
   slopcamera direct assemble <direct-id> [--json]
 
 Init writes a new recipe without overwriting. Edit its ID, shots, prompts and settings.
-Anchor imports one explicit local image and returns a hash-bound source reference.
+Anchor imports one explicit local image or video and returns a hash-bound source reference.
 Plan checks live Gateway video capabilities and duration pricing without a paid call.
 Start retains one budget for the direct ID; starting again cannot reset it.
-Generate dispatches at most one new take and requires a unique take_<id>. Local images,
+Generate dispatches at most one new take and requires a unique take_<id>. Local media,
 including a predecessor's endpoint, require --allow-cloud-upload on that invocation.
 URL-only models require --allow-reference-hosting and a private Vercel Blob store.
-Only exact reference images receive short-lived signed read URLs. Signed URLs and
+Only exact shot references receive short-lived signed read URLs. Signed URLs and
 credentials are not saved. Cleanup deletes those exact objects; inspect cleanup
 receipts and use direct cleanup after interruptions. Blob charges are separate
 from the model catalog estimate and require room in your overall spending limit.

package/apps/desktop/cli/main.ts

@@ -55,7 +55,7 @@ export async function main(
   }
   const paths = await resolveRepositoryPaths(processIo.cwd(), processIo.env);
   if (earlyCommand.kind === "html-render" || earlyCommand.kind === "spatial-world" || earlyCommand.kind === "spatial-asset" || earlyCommand.kind === "spatial-design" || earlyCommand.kind === "directing" || earlyCommand.kind === "studio"
-    || earlyCommand.kind === "spatial-scene" && (earlyCommand.action === "camera-track" || earlyCommand.action === "render-audit" || earlyCommand.action === "review")) {
+    || earlyCommand.kind === "spatial-scene" && (earlyCommand.action === "render" || earlyCommand.action === "camera-track" || earlyCommand.action === "render-audit" || earlyCommand.action === "review")) {
     const controller = new AbortController();
     const cancel = () => controller.abort(new CliError("cancelled", "Command interrupted."));
     process.on("SIGINT", cancel);

package/apps/desktop/cli/project-renderer.ts

@@ -959,6 +959,7 @@ function overlayVideoChain(
   } else {
     filters.push(
       `trim=duration=${seconds(sliceDurationUs)}`,
+      "settb=AVTB",
       `setpts=PTS-STARTPTS+${seconds(resolved.outputRange.startUs)}/TB`,
     );
   }
@@ -1246,7 +1247,7 @@ export async function buildProjectFfmpegInvocation(
     if (overlayInputIndex.has(overlay.overlayId)) continue;
     overlayInputIndex.set(overlay.overlayId, mediaInputIndex.size + overlayInputIndex.size);
     if (overlay.source.kind === "image" || overlay.source.kind === "svg" || overlay.source.kind === "emoji") {
-      inputArguments.push("-loop", "1");
+      inputArguments.push("-loop", "1", "-framerate", emittedRate);
     }
     const prepared = await prepareProjectOverlayMedia(overlay, options);
     preparedOverlayMedia.set(overlay.overlayId, prepared);
@@ -1309,6 +1310,7 @@ export async function buildProjectFfmpegInvocation(
     } else {
       const transparent = `video_blend_canvas_${serial++}`;
       const positioned = `video_blend_positioned_${serial++}`;
+      const layerColorSource = `video_blend_color_source_${serial++}`;
       const layerColor = `video_blend_color_${serial++}`;
       const layerAlphaSource = `video_blend_alpha_source_${serial++}`;
       const layerMask = `video_blend_mask_${serial++}`;
@@ -1317,10 +1319,13 @@ export async function buildProjectFfmpegInvocation(
       const blended = `video_blend_result_${serial++}`;
       filters.push(
         `color=c=black@0:s=${plan.output.pixelWidth}x${plan.output.pixelHeight}:r=${emittedRate}:d=${seconds(plan.output.durationUs)},format=rgba[${transparent}]`,
-        `[${transparent}][${label}]overlay=x=${transform.x}:y=${transform.y}:eof_action=repeat:repeatlast=1:enable='${enable}'[${positioned}]`,
-        `[${positioned}]format=rgba,split=2[${layerColor}][${layerAlphaSource}]`,
+        // Preserve the positioned layer's alpha and blend color channels in RGB.
+        // YUV blend arithmetic changes even uncovered pixels (neutral chroma is 128).
+        `[${transparent}][${label}]overlay=x=${transform.x}:y=${transform.y}:eof_action=repeat:repeatlast=1:enable='${enable}':format=auto[${positioned}]`,
+        `[${positioned}]format=rgba,split=2[${layerColorSource}][${layerAlphaSource}]`,
+        `[${layerColorSource}]format=gbrp[${layerColor}]`,
         `[${layerAlphaSource}]alphaextract[${layerMask}]`,
-        `[${currentVideo}]split=2[${baseBlend}][${baseMerge}]`,
+        `[${currentVideo}]format=gbrp,split=2[${baseBlend}][${baseMerge}]`,
         `[${baseBlend}][${layerColor}]blend=all_mode=${slice.presentation.blendMode}[${blended}]`,
         `[${baseMerge}][${blended}][${layerMask}]maskedmerge[${next}]`,
       );
@@ -1365,15 +1370,23 @@ export async function buildProjectFfmpegInvocation(
     const x = overlayPositionExpression(overlay, "x", bounds.startUs, visibleFullEndUs, animations.entranceUs, animations.exitUs);
     const y = overlayPositionExpression(overlay, "y", bounds.startUs, visibleFullEndUs, animations.entranceUs, animations.exitUs);
     const next = `canvas_${serial++}`;
-    const enable = `between(t,${seconds(resolvedOverlay.outputRange.startUs)},${seconds(chain.visibleEndUs)})`;
+    const staticSource = source.kind === "image" || source.kind === "svg" || source.kind === "emoji";
+    // A still's last decoded PTS can precede the final output frame, especially
+    // after cuts or speed edits. Hold it only inside its half-open interval;
+    // animated sources retain their explicit hide/freeze/loop behavior.
+    const enable = staticSource
+      ? `gte(t,${seconds(resolvedOverlay.outputRange.startUs)})*lt(t,${seconds(chain.visibleEndUs)})`
+      : `between(t,${seconds(resolvedOverlay.outputRange.startUs)},${seconds(chain.visibleEndUs)})`;
+    const terminal = staticSource ? "eof_action=repeat:repeatlast=1" : "eof_action=pass:repeatlast=0";
     const blendMode = overlay.blendMode ?? "normal";
     if (blendMode === "normal") {
       filters.push(
-        `[${currentVideo}][${label}]overlay=x='${x}':y='${y}':eof_action=pass:repeatlast=0:enable='${enable}'[${next}]`,
+        `[${currentVideo}][${label}]overlay=x='${x}':y='${y}':${terminal}:enable='${enable}'[${next}]`,
       );
     } else {
       const transparent = `graphic_blend_canvas_${serial++}`;
       const positioned = `graphic_blend_positioned_${serial++}`;
+      const layerColorSource = `graphic_blend_color_source_${serial++}`;
       const layerColor = `graphic_blend_color_${serial++}`;
       const layerAlphaSource = `graphic_blend_alpha_source_${serial++}`;
       const layerMask = `graphic_blend_mask_${serial++}`;
@@ -1382,10 +1395,11 @@ export async function buildProjectFfmpegInvocation(
       const blended = `graphic_blend_result_${serial++}`;
       filters.push(
         `color=c=black@0:s=${plan.output.pixelWidth}x${plan.output.pixelHeight}:r=${emittedRate}:d=${seconds(plan.output.durationUs)},format=rgba[${transparent}]`,
-        `[${transparent}][${label}]overlay=x='${x}':y='${y}':eof_action=pass:repeatlast=0:enable='${enable}'[${positioned}]`,
-        `[${positioned}]format=rgba,split=2[${layerColor}][${layerAlphaSource}]`,
+        `[${transparent}][${label}]overlay=x='${x}':y='${y}':${terminal}:enable='${enable}':format=auto[${positioned}]`,
+        `[${positioned}]format=rgba,split=2[${layerColorSource}][${layerAlphaSource}]`,
+        `[${layerColorSource}]format=gbrp[${layerColor}]`,
         `[${layerAlphaSource}]alphaextract[${layerMask}]`,
-        `[${currentVideo}]split=2[${baseBlend}][${baseMerge}]`,
+        `[${currentVideo}]format=gbrp,split=2[${baseBlend}][${baseMerge}]`,
         `[${baseBlend}][${layerColor}]blend=all_mode=${blendMode}[${blended}]`,
         `[${baseMerge}][${blended}][${layerMask}]maskedmerge[${next}]`,
       );

package/apps/desktop/cli/spatial-scene-service.ts

@@ -337,7 +337,7 @@ export async function executeSpatialSceneCommand(application: ApplicationContext
   if (command.action === "plan" || command.action === "render") {
     const request = bindSpatialCliExecutionProfile(await readSpatialJson(resolve(application.paths.repositoryRoot, command.request)), command.executionProfile);
     if (command.action === "plan") return planSpatialRender(scene, request);
-    const result = await createApplicationOperationRegistry().execute({ application, abortSignal: new AbortController().signal }, {
+    const result = await createApplicationOperationRegistry().execute({ application, abortSignal: signal ?? new AbortController().signal }, {
       kind: "scene.render", version: 1, input: {
         source: { path: relative(application.paths.repositoryRoot, sourcePath) }, request,
         ...(command.assets === undefined ? {} : { assets: await readSpatialJson(resolve(application.paths.repositoryRoot, command.assets)) }),