vidspotai-shared 1.0.124 → 1.0.126

package/lib/services/agent/executor/visual.js

@@ -5,6 +5,7 @@ const logger_1 = require("../../../utils/logger");
 const referenceImageRenderer_1 = require("../referenceImageRenderer");
 const extendChain_1 = require("../extendChain");
 const helpers_1 = require("../../aiGen/helpers");
+const chains_1 = require("../providerFallback/chains");
 const unsplash_service_1 = require("../../aiGen/providers/unsplash/unsplash.service");
 const duration_1 = require("./duration");
 async function runVisual(deps, scene, ctx, aspect, plan) {
@@ -33,6 +34,17 @@ async function runVisual(deps, scene, ctx, aspect, plan) {
     const identityRefs = anchorImageUrl
         ? Array.from(new Set([anchorImageUrl, ...referenceImageUrls]))
         : referenceImageUrls;
+    // Video models render 9:16 / 16:9 / 1:1 only; a 4:5 plan gets the closest.
+    const videoAspect = (0, chains_1.videoAspectFor)(aspect);
+    // The scene's camera move (from a @-mentioned motion ref or the planner).
+    const cameraControl = scene.cameraPreset ? { preset: scene.cameraPreset } : undefined;
+    // Cache identity for the two fields above, added only when they differ from
+    // what every pre-existing cache entry was minted with (9:16, no camera) so a
+    // re-execute of an older run still replays instead of re-submitting.
+    const videoCacheExtras = {
+        ...(videoAspect !== "9:16" ? { videoAspect } : {}),
+        ...(cameraControl ? { cameraPreset: scene.cameraPreset } : {}),
+    };
     const wrap = (tool, outcome) => outcome.ok
         ? { ok: true, output: { tool, ...outcome.output } }
         : { ok: false, error: outcome.error };
@@ -155,6 +167,7 @@ async function runVisual(deps, scene, ctx, aspect, plan) {
                 referenceImageUrls,
                 extendSegments: extendChain?.length ?? 0,
                 ...(useFlf ? { lastFramePrompt: scene.lastFramePrompt } : {}),
+                ...videoCacheExtras,
             };
             const { hit, inputsHash } = deps.checkTaskCache(clipId, "animate_image", cacheInput);
             if (hit) {
@@ -176,7 +189,8 @@ async function runVisual(deps, scene, ctx, aspect, plan) {
             const img = await deps.runTool("generate_image", {
                 prompt: scene.prompt,
                 modelKey: "google-nano-banana",
-                aspectRatio: aspect,
+                // The still is the clip's first frame — same shape as the clip.
+                aspectRatio: videoAspect,
                 ...(identityRefs.length
                     ? { inputImageUrls: identityRefs }
                     : {}),
@@ -193,7 +207,7 @@ async function runVisual(deps, scene, ctx, aspect, plan) {
                 const endImg = await deps.runTool("generate_image", {
                     prompt: scene.lastFramePrompt,
                     modelKey: "google-nano-banana",
-                    aspectRatio: aspect,
+                    aspectRatio: videoAspect,
                     ...(identityRefs.length
                         ? { inputImageUrls: identityRefs }
                         : {}),
@@ -207,6 +221,10 @@ async function runVisual(deps, scene, ctx, aspect, plan) {
                 motionHint: scene.prompt,
                 durationSec: animateDurationSec,
                 modelKey: animateModelKey,
+                // Was never passed, so every agent i2v clip rendered at the tool's
+                // 9:16 default regardless of the plan's aspect.
+                aspectRatio: videoAspect,
+                ...(cameraControl ? { cameraControl } : {}),
                 ...(lastFrameImageUrl ? { lastFrameImageUrl } : {}),
             }, ctx);
             if (!animate.ok)
@@ -232,6 +250,20 @@ async function runVisual(deps, scene, ctx, aspect, plan) {
         }
         case "ai-text-to-video": {
             const t2vModelKey = scene.preferredModel ?? "kling-v2.6";
+            // A pinned subject (a @-mentioned character / product, or an earlier
+            // scene's identity still) has to reach the model as an image. Text-to-
+            // video models that take reference images get them directly below; for
+            // one that doesn't, the scene goes through image-to-video instead — a
+            // first frame generated WITH the refs, then animated — rather than a
+            // prompt-only clip of a stranger.
+            if (identityRefs.length && (0, chains_1.modelReferenceImageLimit)(t2vModelKey) === 0) {
+                logger_1.logger.info("executor: t2v scene has reference images but its model takes none — using image-to-video", {
+                    sceneIndex: scene.sceneIndex,
+                    modelKey: t2vModelKey,
+                    refs: identityRefs.length,
+                });
+                return runVisual(deps, { ...scene, strategy: "ai-image-to-video" }, ctx, aspect, plan);
+            }
             // AG-2 / L-4: same extend-chain decision as ai-image-to-video.
             const chain = (0, extendChain_1.planExtendChain)(t2vModelKey, durationSec);
             const extendChain = chain?.segments;
@@ -254,6 +286,9 @@ async function runVisual(deps, scene, ctx, aspect, plan) {
                 modelKey: t2vModelKey,
                 durationSec: t2vDurationSec,
                 extendSegments: extendChain?.length ?? 0,
+                ...videoCacheExtras,
+                // Bible refs, not identityRefs — same reasoning as ai-image-to-video.
+                ...(identityRefs.length ? { referenceImageUrls } : {}),
             };
             const { hit, inputsHash } = deps.checkTaskCache(clipId, "generate_video", cacheInput);
             if (hit) {
@@ -273,6 +308,9 @@ async function runVisual(deps, scene, ctx, aspect, plan) {
                 prompt: scene.prompt,
                 modelKey: t2vModelKey,
                 durationSec: t2vDurationSec,
+                aspectRatio: videoAspect,
+                ...(cameraControl ? { cameraControl } : {}),
+                ...(identityRefs.length ? { referenceImageUrls: identityRefs.slice(0, 9) } : {}),
             }, ctx);
             if (!out.ok)
                 return wrap("generate_video", out);

package/lib/services/agent/mentionedRefs.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"mentionedRefs.d.ts","sourceRoot":"","sources":["../../../src/services/agent/mentionedRefs.ts"],"names":[],"mappings":"AAAA,OAAO,EAAgB,UAAU,EAAE,MAAM,4BAA4B,CAAC;AACtE,OAAO,EAAE,gBAAgB,EAAe,MAAM,gCAAgC,CAAC;AAgD/E;;;;;;;;;;GAUG;AACH,wBAAgB,yBAAyB,CACvC,KAAK,EAAE,gBAAgB,GAAG,SAAS,EACnC,KAAK,EAAE,UAAU,GAChB,gBAAgB,GAAG,SAAS,CAwC9B;AAID;;;GAGG;AACH,wBAAgB,yBAAyB,CACvC,KAAK,EAAE,UAAU,EACjB,KAAK,EAAE,gBAAgB,GAAG,IAAI,GAC7B,MAAM,CAmER"}
+{"version":3,"file":"mentionedRefs.d.ts","sourceRoot":"","sources":["../../../src/services/agent/mentionedRefs.ts"],"names":[],"mappings":"AAAA,OAAO,EAAgB,UAAU,EAAE,MAAM,4BAA4B,CAAC;AACtE,OAAO,EAAE,gBAAgB,EAAe,MAAM,gCAAgC,CAAC;AAkD/E;;;;;;;;;;GAUG;AACH,wBAAgB,yBAAyB,CACvC,KAAK,EAAE,gBAAgB,GAAG,SAAS,EACnC,KAAK,EAAE,UAAU,GAChB,gBAAgB,GAAG,SAAS,CAwC9B;AAID;;;GAGG;AACH,wBAAgB,yBAAyB,CACvC,KAAK,EAAE,UAAU,EACjB,KAAK,EAAE,gBAAgB,GAAG,IAAI,GAC7B,MAAM,CA2ER"}

package/lib/services/agent/mentionedRefs.js

@@ -123,7 +123,10 @@ function buildMentionedRefsSection(brief, bible) {
     const moves = by("motion");
     if (moves.length) {
         lines.push("  Camera moves — same placement rule as angles. Copy the direction into scene.prompt verbatim.", "  A still can't carry a camera move, so those scenes must use ai-image-to-video or", "  ai-text-to-video (never ai-image-static / stock-image-ken-burns):");
-        moves.forEach((r) => lines.push(`    ${r.name}: ${r.instruction}`));
+        moves.forEach((r) => lines.push(`    ${r.name}: ${r.instruction}${r.presetKey ? ` [cameraPreset: ${r.presetKey}]` : ""}`));
+        if (moves.some((r) => r.presetKey)) {
+            lines.push("  For a move tagged [cameraPreset: X], also set scene.cameraPreset = X on those scenes — the", "  video model gets it as a structured camera instruction in the syntax it understands.");
+        }
     }
     return lines.join("\n");
 }

package/lib/services/agent/planner/structuralRules.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"structuralRules.d.ts","sourceRoot":"","sources":["../../../../src/services/agent/planner/structuralRules.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAoB,MAAM,+BAA+B,CAAC;AAG7E;;;;;;;;;;;;;;;;GAgBG;AACH,wBAAgB,2BAA2B,CAAC,KAAK,EAAE,UAAU,GAAG,MAAM,CA0OrE"}
+{"version":3,"file":"structuralRules.d.ts","sourceRoot":"","sources":["../../../../src/services/agent/planner/structuralRules.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAoB,MAAM,+BAA+B,CAAC;AAG7E;;;;;;;;;;;;;;;;GAgBG;AACH,wBAAgB,2BAA2B,CAAC,KAAK,EAAE,UAAU,GAAG,MAAM,CAkPrE"}

package/lib/services/agent/planner/structuralRules.js

@@ -130,12 +130,13 @@ function buildStructuralRulesSection(brief) {
     //   • 3s floor: kling-v3.0 (T3)
     //   • 4s floor: google-veo-3.1 / -fast / -lite, bytedance-seedance-1.5-pro,
     //     bytedance-seedance-2.0 / -2.0-fast
-    //   • 5s floor: kling 2.x family, pixverse all, alibaba-wan, runway-gen4-turbo
+    //   • 5s floor: kling 2.x family, pixverse all, alibaba-wan
+    //   (runway-gen4-turbo now takes 2-10s, like gen4.5)
     //   • 6s floor: minimax-hailuo-02 / 2.3 / 2.3-fast  ← do NOT pick for <6s scenes.
     // The catalog above lists each model's "dur:..." constraint — match scene
     // durations to it. Pacing target can still be honoured by varying scene COUNT
     // (e.g. a 10s UGC at fast pacing = 2 scenes × 5s on pixverse, NOT 6 × 1.7s).
-    lines.push("  - Provider duration floor: every scene with strategy=ai-image-to-video or ai-text-to-video MUST have durationMs/1000 in its preferredModel's allowed durations (shown as `dur:` in the catalog). Hard universal floor is 2s — no provider accepts sub-2s clips. If a desired scene is shorter than the floor, MERGE it with the next scene instead of generating a too-short clip. For sub-5s scenes, pick from: bytedance-seedance-1.0-pro / -pro-fast (2-12s), runway-gen4.5 (2-10s), kling-v3.0 (3-15s), google-veo-3.1* (4-8s). Picking minimax-hailuo for a <6s scene is a HARD ERROR.");
+    lines.push("  - Provider duration floor: every scene with strategy=ai-image-to-video or ai-text-to-video MUST have durationMs/1000 in its preferredModel's allowed durations (shown as `dur:` in the catalog). Hard universal floor is 2s — no provider accepts sub-2s clips. If a desired scene is shorter than the floor, MERGE it with the next scene instead of generating a too-short clip. For sub-5s scenes, pick from: bytedance-seedance-1.0-pro / -pro-fast (2-12s), runway-gen4.5 / runway-gen4-turbo (2-10s), kling-v3.0 (3-15s), google-veo-3.1* (4-8s). Picking minimax-hailuo for a <6s scene is a HARD ERROR.");
     // AG-28: voiceover-vs-scene duration budget. Without this, the planner
     // picks scene.durationMs based on visual pacing and the TTS that comes
     // out is consistently 1-2s longer per scene → voice clipped mid-sentence.
@@ -189,7 +190,11 @@ function buildStructuralRulesSection(brief) {
     // Only meaningful on ai-image-to-video scenes whose preferredModel is
     // FLF-capable; the executor silently ignores it otherwise, so the rule is
     // additive and safe to skip when no scene needs directed two-anchor motion.
-    lines.push("  - First-last-frame (FLF) interpolation: for a scene with strategy=ai-image-to-video where the motion goes between TWO distinct, specific compositions (wide establishing shot → product close-up; closed door → open door; ingredient on counter → finished plated dish; logo small → logo filling frame), set scene.lastFramePrompt to a COMPLETE image prompt describing the END frame. scene.prompt is the START frame; the model interpolates between them for reliable directed motion instead of guessing from a single image + motion words. Rules: (1) lastFramePrompt is a full visual description of the final composition, NOT a motion verb (GOOD: \"extreme close-up of the bottle label, soft studio light\"; BAD: \"zoom into the label\"). (2) Keep subject/lighting/style consistent with scene.prompt so the interpolation reads as one continuous shot — re-state shared identity descriptors verbatim. (3) ONLY set it when preferredModel is FLF-capable: google-veo-3.1 / -fast, kling-v2.6 / -v2.5-turbo / -v3.0, runway-gen4-turbo / -gen4.5, alibaba-wan. On any other model it is ignored. (4) Use sparingly — only when a scene genuinely has two anchor compositions; ordinary single-image motion needs no lastFramePrompt. It costs one extra (cheap) image generation per FLF scene.");
+    lines.push("  - First-last-frame (FLF) interpolation: for a scene with strategy=ai-image-to-video where the motion goes between TWO distinct, specific compositions (wide establishing shot → product close-up; closed door → open door; ingredient on counter → finished plated dish; logo small → logo filling frame), set scene.lastFramePrompt to a COMPLETE image prompt describing the END frame. scene.prompt is the START frame; the model interpolates between them for reliable directed motion instead of guessing from a single image + motion words. Rules: (1) lastFramePrompt is a full visual description of the final composition, NOT a motion verb (GOOD: \"extreme close-up of the bottle label, soft studio light\"; BAD: \"zoom into the label\"). (2) Keep subject/lighting/style consistent with scene.prompt so the interpolation reads as one continuous shot — re-state shared identity descriptors verbatim. (3) ONLY set it when preferredModel is FLF-capable: google-veo-3.1 / -fast, kling-v2.6 / -v2.5-turbo / -v3.0, alibaba-wan (NOT Runway — its models take a first frame only). On any other model it is ignored. (4) Use sparingly — only when a scene genuinely has two anchor compositions; ordinary single-image motion needs no lastFramePrompt. It costs one extra (cheap) image generation per FLF scene.");
+    // Camera move as a structured field. The executor sends it to the video
+    // model as cameraControl; applyCameraDirection writes it in the syntax that
+    // model documents (MiniMax [Pan left] commands, a camera sentence, etc.).
+    lines.push("  - Camera move: on an ai-image-to-video or ai-text-to-video scene with ONE clear camera move, set scene.cameraPreset to it — one of static, pan_left, pan_right, tilt_up, tilt_down, zoom_in, zoom_out, dolly_in, dolly_out, orbit_left, orbit_right, roll_left, roll_right, tracking, handheld, push_in, pull_out. Keep describing the move in scene.prompt too. Leave it unset when the shot has no particular camera move or a compound one the list can't name.");
     // text-layout v1 — per-scene composition decision. Optional but strongly
     // recommended when source aspect ≠ target aspect (i.e. most stock video
     // and avatar shots on 9:16 vertical, or vertical UGC on 16:9 horizontal).

package/lib/services/agent/poster/posterAgent.schema.d.ts

@@ -186,8 +186,8 @@ export declare const PosterVariantPlanSchema: z.ZodObject<{
     }, z.core.$strip>;
     scrim: z.ZodEnum<{
         none: "none";
-        soft: "soft";
         strong: "strong";
+        soft: "soft";
     }>;
     imagePrompt: z.ZodString;
     negativePrompt: z.ZodOptional<z.ZodString>;
@@ -297,8 +297,8 @@ export declare const PosterPlanSchema: z.ZodObject<{
         }, z.core.$strip>;
         scrim: z.ZodEnum<{
             none: "none";
-            soft: "soft";
             strong: "strong";
+            soft: "soft";
         }>;
         imagePrompt: z.ZodString;
         negativePrompt: z.ZodOptional<z.ZodString>;
@@ -446,8 +446,8 @@ export declare const PosterRefineRequestSchema: z.ZodObject<{
             }, z.core.$strip>;
             scrim: z.ZodEnum<{
                 none: "none";
-                soft: "soft";
                 strong: "strong";
+                soft: "soft";
             }>;
             imagePrompt: z.ZodString;
             negativePrompt: z.ZodOptional<z.ZodString>;

package/lib/services/agent/providerFallback/chains.d.ts

@@ -52,6 +52,31 @@ export interface VideoCapabilityRequest {
 export declare function isVideoModelCapable(model: EVideoGenModels, req: VideoCapabilityRequest): boolean;
 /** Reduce a "W:H" or "WxH" ratio to lowest terms, e.g. "1280:720" → "16:9". */
 export declare function normalizeAspect(a: string): string;
+/**
+ * How many reference images (identity / style refs that are NOT the first
+ * frame) a video model takes natively: its `referenceImageUrls.max`, Infinity
+ * when it declares the field without a cap, 0 when it doesn't take any.
+ */
+export declare function modelReferenceImageLimit(m: EVideoGenModels | string): number;
+/**
+ * Soft capability ordering for a fallback chain: models that natively cover
+ * every "nice to have" in `wants` go first, the rest keep their relative
+ * order behind them. Used for inputs a model can do without but renders worse
+ * without (reference images → the subject stops looking like the ref), so a
+ * capable model is tried before one that would drop them, without emptying
+ * the chain when none is capable.
+ */
+export interface VideoPreference {
+    /** Reference images the caller wants honored. */
+    referenceImages?: number;
+}
+export declare function rankVideoChainByPreference(chain: readonly EVideoGenModels[], wants: VideoPreference): EVideoGenModels[];
+/**
+ * The closest ratio a video model can render for a plan aspect. Every video
+ * model takes 9:16 / 16:9 / 1:1; 4:5 (feed) is image-only, and 1:1 is the
+ * closest shape, so the render crops the least from it.
+ */
+export declare function videoAspectFor(aspect: string): "9:16" | "16:9" | "1:1";
 /**
  * Build a chain that puts the planner's pick first, then the rest of the
  * tier-appropriate chain (deduped). Honors the planner's intent while still

package/lib/services/agent/providerFallback/chains.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"chains.d.ts","sourceRoot":"","sources":["../../../../src/services/agent/providerFallback/chains.ts"],"names":[],"mappings":"AAAA,OAAO,EAEL,eAAe,EACf,eAAe,EACf,aAAa,EAEb,KAAK,cAAc,EACpB,MAAM,2BAA2B,CAAC;AAGnC;;;;;;;;;;;;GAYG;AAGH,eAAO,MAAM,sBAAsB,EAAE,SAAS,eAAe,EAO5D,CAAC;AAKF,eAAO,MAAM,sBAAsB,EAAE,SAAS,eAAe,EAU5D,CAAC;AAQF,eAAO,MAAM,uBAAuB,EAAE,SAAS,eAAe,EAU7D,CAAC;AAKF,eAAO,MAAM,uBAAuB,EAAE,SAAS,eAAe,EAO7D,CAAC;AASF,eAAO,MAAM,kBAAkB,EAAE,SAAS,eAAe,EAIxD,CAAC;AAGF,eAAO,MAAM,WAAW,EAAE,SAAS,eAAe,EAGjD,CAAC;AAGF,eAAO,MAAM,SAAS,EAAE,SAAS,aAAa,EAE7C,CAAC;AAEF;;;;;;;;;;GAUG;AACH,MAAM,WAAW,sBAAsB;IACrC,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,EAAE,MAAM,CAAC;IACpB,qDAAqD;IACrD,IAAI,CAAC,EAAE,cAAc,CAAC;IACtB,+EAA+E;IAC/E,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,wCAAwC;IACxC,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB;;;;;;;OAOG;IACH,aAAa,CAAC,EAAE,OAAO,CAAC;CACzB;AAED,wBAAgB,mBAAmB,CACjC,KAAK,EAAE,eAAe,EACtB,GAAG,EAAE,sBAAsB,GAC1B,OAAO,CAuCT;AAED,+EAA+E;AAC/E,wBAAgB,eAAe,CAAC,CAAC,EAAE,MAAM,GAAG,MAAM,CASjD;AA+CD;;;;GAIG;AACH,wBAAgB,eAAe,CAC7B,WAAW,EAAE,eAAe,EAC5B,aAAa,EAAE,SAAS,eAAe,EAAE,GACxC,eAAe,EAAE,CAMnB;AAED;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AACH,MAAM,WAAW,mBAAmB;IAClC,4CAA4C;IAC5C,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,wEAAwE;IACxE,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AAED,MAAM,WAAW,kBAAkB;IACjC,8CAA8C;IAC9C,YAAY,EAAE,MAAM,CAAC;IACrB,wEAAwE;IACxE,UAAU,EAAE,MAAM,CAAC;IACnB,yDAAyD;IACzD,QAAQ,EAAE,MAAM,CAAC;IACjB,2EAA2E;IAC3E,aAAa,EAAE,OAAO,CAAC;CACxB;AAED,wBAAgB,oBAAoB,CAClC,KAAK,EAAE,MAAM,GAAG,eAAe,EAC/B,WAAW,EAAE,MAAM,EACnB,IAAI,GAAE,mBAAwB,GAC7B,kBAAkB,GAAG,IAAI,CAyB3B"}
+{"version":3,"file":"chains.d.ts","sourceRoot":"","sources":["../../../../src/services/agent/providerFallback/chains.ts"],"names":[],"mappings":"AAAA,OAAO,EAEL,eAAe,EACf,eAAe,EACf,aAAa,EAEb,KAAK,cAAc,EACpB,MAAM,2BAA2B,CAAC;AAGnC;;;;;;;;;;;;GAYG;AAGH,eAAO,MAAM,sBAAsB,EAAE,SAAS,eAAe,EAY5D,CAAC;AAKF,eAAO,MAAM,sBAAsB,EAAE,SAAS,eAAe,EAW5D,CAAC;AAQF,eAAO,MAAM,uBAAuB,EAAE,SAAS,eAAe,EAW7D,CAAC;AAKF,eAAO,MAAM,uBAAuB,EAAE,SAAS,eAAe,EAU7D,CAAC;AASF,eAAO,MAAM,kBAAkB,EAAE,SAAS,eAAe,EAIxD,CAAC;AAGF,eAAO,MAAM,WAAW,EAAE,SAAS,eAAe,EAGjD,CAAC;AAGF,eAAO,MAAM,SAAS,EAAE,SAAS,aAAa,EAE7C,CAAC;AAEF;;;;;;;;;;GAUG;AACH,MAAM,WAAW,sBAAsB;IACrC,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,EAAE,MAAM,CAAC;IACpB,qDAAqD;IACrD,IAAI,CAAC,EAAE,cAAc,CAAC;IACtB,+EAA+E;IAC/E,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,wCAAwC;IACxC,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB;;;;;;;OAOG;IACH,aAAa,CAAC,EAAE,OAAO,CAAC;CACzB;AAED,wBAAgB,mBAAmB,CACjC,KAAK,EAAE,eAAe,EACtB,GAAG,EAAE,sBAAsB,GAC1B,OAAO,CAuCT;AAED,+EAA+E;AAC/E,wBAAgB,eAAe,CAAC,CAAC,EAAE,MAAM,GAAG,MAAM,CASjD;AA6DD;;;;GAIG;AACH,wBAAgB,wBAAwB,CAAC,CAAC,EAAE,eAAe,GAAG,MAAM,GAAG,MAAM,CAI5E;AAED;;;;;;;GAOG;AACH,MAAM,WAAW,eAAe;IAC9B,iDAAiD;IACjD,eAAe,CAAC,EAAE,MAAM,CAAC;CAC1B;AAED,wBAAgB,0BAA0B,CACxC,KAAK,EAAE,SAAS,eAAe,EAAE,EACjC,KAAK,EAAE,eAAe,GACrB,eAAe,EAAE,CAInB;AAED;;;;GAIG;AACH,wBAAgB,cAAc,CAAC,MAAM,EAAE,MAAM,GAAG,MAAM,GAAG,MAAM,GAAG,KAAK,CAOtE;AAED;;;;GAIG;AACH,wBAAgB,eAAe,CAC7B,WAAW,EAAE,eAAe,EAC5B,aAAa,EAAE,SAAS,eAAe,EAAE,GACxC,eAAe,EAAE,CAMnB;AAED;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AACH,MAAM,WAAW,mBAAmB;IAClC,4CAA4C;IAC5C,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,wEAAwE;IACxE,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AAED,MAAM,WAAW,kBAAkB;IACjC,8CAA8C;IAC9C,YAAY,EAAE,MAAM,CAAC;IACrB,wEAAwE;IACxE,UAAU,EAAE,MAAM,CAAC;IACnB,yDAAyD;IACzD,QAAQ,EAAE,MAAM,CAAC;IACjB,2EAA2E;IAC3E,aAAa,EAAE,OAAO,CAAC;CACxB;AAED,wBAAgB,oBAAoB,CAClC,KAAK,EAAE,MAAM,GAAG,eAAe,EAC/B,WAAW,EAAE,MAAM,EACnB,IAAI,GAAE,mBAAwB,GAC7B,kBAAkB,GAAG,IAAI,CAyB3B"}

package/lib/services/agent/providerFallback/chains.js

@@ -3,6 +3,9 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.SFX_CHAIN = exports.MUSIC_CHAIN = exports.AVATAR_VIDEO_CHAIN = exports.T3_IMAGE_TO_VIDEO_CHAIN = exports.T2_IMAGE_TO_VIDEO_CHAIN = exports.T2_TEXT_TO_VIDEO_CHAIN = exports.T3_TEXT_TO_VIDEO_CHAIN = void 0;
 exports.isVideoModelCapable = isVideoModelCapable;
 exports.normalizeAspect = normalizeAspect;
+exports.modelReferenceImageLimit = modelReferenceImageLimit;
+exports.rankVideoChainByPreference = rankVideoChainByPreference;
+exports.videoAspectFor = videoAspectFor;
 exports.buildVideoChain = buildVideoChain;
 exports.snapDurationForModel = snapDurationForModel;
 const aiModels_1 = require("../../../globals/aiModels");
@@ -23,11 +26,16 @@ const helpers_1 = require("../../aiGen/helpers");
 // ── Video: text-to-video, T3 (cinematic) ──────────────────────────────
 exports.T3_TEXT_TO_VIDEO_CHAIN = [
     aiModels_1.EVideoGenModels.GOOGLE_VEO_3_1,
+    // Gemini Omni Flash: Google's current default video model (GA), 3-10s, ~$0.10/s.
+    aiModels_1.EVideoGenModels.GOOGLE_GEMINI_OMNI_FLASH,
     aiModels_1.EVideoGenModels.KLING_V2_6,
     aiModels_1.EVideoGenModels.RUNWAY_GEN4_5,
     aiModels_1.EVideoGenModels.BYTEDANCE_SEEDANCE_2_5,
     aiModels_1.EVideoGenModels.BYTEDANCE_SEEDANCE_2_0,
+    aiModels_1.EVideoGenModels.MINIMAX_H3,
     aiModels_1.EVideoGenModels.MINIMAX_HAILUO_2_3,
+    // PixVerse C1: cinema-grade flagship, any 1–15s at up to 1080p + native audio.
+    aiModels_1.EVideoGenModels.PIXVERSE_C1,
 ];
 // ── Video: text-to-video, T2 (balanced) ───────────────────────────────
 // AG-16: seedance-1.0-pro-fast added to cover 2-3s clips; pixverse-v6 added
@@ -41,6 +49,7 @@ exports.T2_TEXT_TO_VIDEO_CHAIN = [
     aiModels_1.EVideoGenModels.BYTEDANCE_SEEDANCE_2_0_MINI,
     aiModels_1.EVideoGenModels.PIXVERSE_V6,
     aiModels_1.EVideoGenModels.PIXVERSE_V5_6,
+    aiModels_1.EVideoGenModels.MINIMAX_H3_MAX,
     aiModels_1.EVideoGenModels.MINIMAX_HAILUO_2_3_FAST,
 ];
 // ── Video: image-to-video, T2 (most common path) ──────────────────────
@@ -58,6 +67,7 @@ exports.T2_IMAGE_TO_VIDEO_CHAIN = [
     // 2.0 Mini: cheapest 2.x I2V (native audio + refs, up to 15s). Budget tier → T2.
     aiModels_1.EVideoGenModels.BYTEDANCE_SEEDANCE_2_0_MINI,
     aiModels_1.EVideoGenModels.PIXVERSE_V6,
+    aiModels_1.EVideoGenModels.MINIMAX_H3_MAX,
     aiModels_1.EVideoGenModels.MINIMAX_HAILUO_02,
 ];
 // ── Video: image-to-video, T3 (premium) ───────────────────────────────
@@ -67,9 +77,12 @@ exports.T3_IMAGE_TO_VIDEO_CHAIN = [
     aiModels_1.EVideoGenModels.KLING_V2_6,
     aiModels_1.EVideoGenModels.KLING_V3_0,
     aiModels_1.EVideoGenModels.GOOGLE_VEO_3_1,
+    aiModels_1.EVideoGenModels.GOOGLE_GEMINI_OMNI_FLASH,
     aiModels_1.EVideoGenModels.RUNWAY_GEN4_5,
     aiModels_1.EVideoGenModels.BYTEDANCE_SEEDANCE_2_5,
     aiModels_1.EVideoGenModels.BYTEDANCE_SEEDANCE_2_0,
+    // PixVerse C1: cinema-grade flagship, any 1–15s at up to 1080p + native audio.
+    aiModels_1.EVideoGenModels.PIXVERSE_C1,
 ];
 // ── Video: avatar / talking-head ──────────────────────────────────────
 // Kling (byo-media) is the default; HeyGen/Tavus (hosted) are fallbacks. NOTE:
@@ -170,11 +183,15 @@ function gcd(a, b) {
 //   EVideoGenModels.BYTEDANCE_SEEDANCE_2_0,
 //   EVideoGenModels.PIXVERSE_V5_6,
 // ]);
-// AG-16: small allow-list for native-audio support — not yet on the schema as a
-// queryable field, so kept hardcoded (conservative). Extend when audio becomes
-// a proper config field like lastFrameImageUrl did.
+// AG-16: native-audio support. A model that declares `audio` in its config
+// fields supports it (Veo 3.x, Seedance 1.5-pro / 2.x, Kling 2.6 / 3.0,
+// Pixverse); this list only covers models whose audio is always on and so
+// declare no toggle (Kling 3.0 Turbo, the avatar model).
 const AUDIO_MODELS = new Set([
-    aiModels_1.EVideoGenModels.GOOGLE_VEO_3,
+    aiModels_1.EVideoGenModels.KLING_V3_0_TURBO,
+    aiModels_1.EVideoGenModels.GOOGLE_GEMINI_OMNI_FLASH,
+    // REMOVED FOR v2 — kept for reference: Veo 3.0 shut down 2026-06-30.
+    // EVideoGenModels.GOOGLE_VEO_3,
     aiModels_1.EVideoGenModels.GOOGLE_VEO_3_1,
     aiModels_1.EVideoGenModels.GOOGLE_VEO_3_1_FAST,
     aiModels_1.EVideoGenModels.GOOGLE_VEO_3_1_LITE,
@@ -182,13 +199,50 @@ const AUDIO_MODELS = new Set([
     aiModels_1.EVideoGenModels.BYTEDANCE_SEEDANCE_2_0,
     aiModels_1.EVideoGenModels.BYTEDANCE_SEEDANCE_2_0_MINI,
     aiModels_1.EVideoGenModels.KLING_AVATAR,
+    aiModels_1.EVideoGenModels.MINIMAX_H3,
+    aiModels_1.EVideoGenModels.MINIMAX_H3_MAX,
+    aiModels_1.EVideoGenModels.PIXVERSE_C1,
 ]);
 function supportsEndFrame(m) {
     // Single source of truth: model config declares `lastFrameImageUrl` in fields.
     return (0, helpers_1.modelSupportsLastFrame)(m);
 }
 function supportsAudio(m) {
-    return AUDIO_MODELS.has(m);
+    return AUDIO_MODELS.has(m) || "audio" in (videoFields(m) ?? {});
+}
+function videoFields(m) {
+    const cfg = aiModels_1.aiModelConfigs[m];
+    return cfg?.fields;
+}
+/**
+ * How many reference images (identity / style refs that are NOT the first
+ * frame) a video model takes natively: its `referenceImageUrls.max`, Infinity
+ * when it declares the field without a cap, 0 when it doesn't take any.
+ */
+function modelReferenceImageLimit(m) {
+    const field = videoFields(m)?.referenceImageUrls;
+    if (!field)
+        return 0;
+    return typeof field.max === "number" ? field.max : Number.POSITIVE_INFINITY;
+}
+function rankVideoChainByPreference(chain, wants) {
+    const satisfies = (m) => !wants.referenceImages || modelReferenceImageLimit(m) > 0;
+    return [...chain.filter(satisfies), ...chain.filter((m) => !satisfies(m))];
+}
+/**
+ * The closest ratio a video model can render for a plan aspect. Every video
+ * model takes 9:16 / 16:9 / 1:1; 4:5 (feed) is image-only, and 1:1 is the
+ * closest shape, so the render crops the least from it.
+ */
+function videoAspectFor(aspect) {
+    if (aspect === "9:16" || aspect === "16:9" || aspect === "1:1")
+        return aspect;
+    const m = aspect.match(/^(\d+):(\d+)$/);
+    if (!m)
+        return "9:16";
+    const r = Number(m[1]) / Number(m[2]);
+    const options = [["9:16", 9 / 16], ["1:1", 1], ["16:9", 16 / 9]];
+    return options.reduce((best, o) => (Math.abs(o[1] - r) < Math.abs(best[1] - r) ? o : best))[0];
 }
 /**
  * Build a chain that puts the planner's pick first, then the rest of the

package/lib/services/agent/repair/strategies/llmRepair.strategy.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"llmRepair.strategy.d.ts","sourceRoot":"","sources":["../../../../../src/services/agent/repair/strategies/llmRepair.strategy.ts"],"names":[],"mappings":"AAOA,OAAO,EAAiC,cAAc,EAAE,MAAM,UAAU,CAAC;AAwOzE,2EAA2E;AAC3E,eAAO,MAAM,uBAAuB,gBAAsC,CAAC;AAC3E,gEAAgE;AAChE,eAAO,MAAM,qBAAqB,gBAAoC,CAAC"}
+{"version":3,"file":"llmRepair.strategy.d.ts","sourceRoot":"","sources":["../../../../../src/services/agent/repair/strategies/llmRepair.strategy.ts"],"names":[],"mappings":"AAOA,OAAO,EAAiC,cAAc,EAAE,MAAM,UAAU,CAAC;AAyOzE,2EAA2E;AAC3E,eAAO,MAAM,uBAAuB,gBAAsC,CAAC;AAC3E,gEAAgE;AAChE,eAAO,MAAM,qBAAqB,gBAAoC,CAAC"}

package/lib/services/agent/repair/strategies/llmRepair.strategy.js

@@ -69,6 +69,8 @@ function alternativeModels(currentModelKey, limit = 14) {
     for (const key of Object.keys(aiModels_1.aiModelConfigs)) {
         if (key === currentModelKey)
             continue;
+        if (!(0, aiModels_1.isModelOffered)(aiModels_1.aiModelConfigs[key]))
+            continue;
         const summary = summarizeModel(key);
         if (!summary?.type)
             continue;

package/lib/services/agent/tools/animateImage.tool.d.ts

@@ -12,6 +12,37 @@ declare const InputSchema: z.ZodObject<{
         "16:9": "16:9";
     }>>;
     lastFrameImageUrl: z.ZodOptional<z.ZodString>;
+    cameraControl: z.ZodOptional<z.ZodObject<{
+        preset: z.ZodOptional<z.ZodEnum<{
+            static: "static";
+            pan_left: "pan_left";
+            pan_right: "pan_right";
+            tilt_up: "tilt_up";
+            tilt_down: "tilt_down";
+            zoom_in: "zoom_in";
+            zoom_out: "zoom_out";
+            dolly_in: "dolly_in";
+            dolly_out: "dolly_out";
+            orbit_left: "orbit_left";
+            orbit_right: "orbit_right";
+            roll_left: "roll_left";
+            roll_right: "roll_right";
+            tracking: "tracking";
+            handheld: "handheld";
+            push_in: "push_in";
+            pull_out: "pull_out";
+        }>>;
+        vector: z.ZodOptional<z.ZodObject<{
+            horizontal: z.ZodOptional<z.ZodNumber>;
+            vertical: z.ZodOptional<z.ZodNumber>;
+            pan: z.ZodOptional<z.ZodNumber>;
+            tilt: z.ZodOptional<z.ZodNumber>;
+            roll: z.ZodOptional<z.ZodNumber>;
+            zoom: z.ZodOptional<z.ZodNumber>;
+        }, z.core.$strip>>;
+        intensity: z.ZodOptional<z.ZodNumber>;
+    }, z.core.$strip>>;
+    negativePrompt: z.ZodOptional<z.ZodString>;
 }, z.core.$strip>;
 declare const OutputSchema: z.ZodObject<{
     status: z.ZodLiteral<"triggered">;

package/lib/services/agent/tools/animateImage.tool.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"animateImage.tool.d.ts","sourceRoot":"","sources":["../../../../src/services/agent/tools/animateImage.tool.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,eAAe,EAAE,MAAM,2BAA2B,CAAC;AAI5D,OAAO,EAAgB,cAAc,EAAE,MAAM,iBAAiB,CAAC;AAW/D,QAAA,MAAM,WAAW;;;;;;;;;;;iBA6Bf,CAAC;AAEH,QAAA,MAAM,YAAY;;;;;;iBAMhB,CAAC;AAEH,eAAO,MAAM,gBAAgB,EAAE,cAAc,CAC3C,OAAO,WAAW,EAClB,OAAO,YAAY,CA2FpB,CAAC"}
+{"version":3,"file":"animateImage.tool.d.ts","sourceRoot":"","sources":["../../../../src/services/agent/tools/animateImage.tool.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,eAAe,EAAE,MAAM,2BAA2B,CAAC;AAM5D,OAAO,EAAgB,cAAc,EAAE,MAAM,iBAAiB,CAAC;AAW/D,QAAA,MAAM,WAAW;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAkCf,CAAC;AAEH,QAAA,MAAM,YAAY;;;;;;iBAMhB,CAAC;AAEH,eAAO,MAAM,gBAAgB,EAAE,cAAc,CAC3C,OAAO,WAAW,EAClB,OAAO,YAAY,CA+FpB,CAAC"}

package/lib/services/agent/tools/animateImage.tool.js

@@ -5,6 +5,8 @@ const zod_1 = require("zod");
 const aiModels_1 = require("../../../globals/aiModels");
 const logger_1 = require("../../../utils/logger");
 const aiGenFactory_service_1 = require("../../aiGen/aiGenFactory.service");
+const cameraDirection_1 = require("../../aiGen/canonicalAdapters/cameraDirection");
+const videoPlan_schema_1 = require("../../../schemas/videoPlan.schema");
 const rateLimiter_1 = require("../../rateLimiter");
 const toolRegistry_1 = require("../toolRegistry");
 const chains_1 = require("../providerFallback/chains");
@@ -30,8 +32,11 @@ const InputSchema = zod_1.z.object({
         .describe("AG-1: end frame for first-last-frame (FLF) interpolation — animate FROM " +
         "imageUrl (first frame) TO this image (last frame) for directed motion " +
         "between two specific compositions. Only FLF-capable models accept it " +
-        "(kling, runway-gen4-turbo, google-veo-3.1, alibaba-wan); the chain skips " +
+        "(kling, google-veo-3.1, alibaba-wan); the chain skips " +
         "and falls back on models that don't. Omit for plain single-image motion."),
+    cameraControl: videoPlan_schema_1.CameraControlSchema.optional().describe("Camera move (preset like dolly_in / pan_left / orbit_right / static, optional 0..1 intensity). " +
+        "Translated per model: MiniMax [Pan left] commands, a camera sentence in the prompt elsewhere."),
+    negativePrompt: zod_1.z.string().optional(),
 });
 const OutputSchema = zod_1.z.object({
     status: zod_1.z.literal("triggered"),
@@ -87,15 +92,17 @@ exports.animateImageTool = {
                 let heldSlot = true;
                 try {
                     const provider = (0, aiGenFactory_service_1.getAiGenProviderService)(modelKey);
-                    const result = await provider.generateVideo({
+                    const result = await provider.generateVideo((0, cameraDirection_1.applyCameraDirection)({
                         prompt: input.motionHint,
                         modelKey,
                         duration: input.durationSec,
                         aspectRatio: input.aspectRatio,
                         inputImageUrl: input.imageUrl,
                         lastFrameImageUrl: input.lastFrameImageUrl,
+                        negativePrompt: input.negativePrompt,
+                        cameraControl: input.cameraControl,
                         type: "image-to-video",
-                    });
+                    }));
                     const taskId = result.task?.taskId ??
                         result.task?.id ??
                         (typeof result.task === "string" ? result.task : "unknown");

package/lib/services/agent/tools/generateImage.tool.d.ts

@@ -21,16 +21,18 @@ import { ToolDefinition } from "../toolRegistry";
 declare const InputSchema: z.ZodObject<{
     prompt: z.ZodString;
     modelKey: z.ZodDefault<z.ZodEnum<{
-        "google-imagen-4-ultra": "google-imagen-4-ultra";
-        "google-imagen-4": "google-imagen-4";
-        "google-imagen-4-fast": "google-imagen-4-fast";
         "google-nano-banana-pro": "google-nano-banana-pro";
         "google-nano-banana": "google-nano-banana";
+        "google-nano-banana-2": "google-nano-banana-2";
         "openai-gpt-image-2": "openai-gpt-image-2";
         "openai-gpt-image-1.5": "openai-gpt-image-1.5";
         "openai-gpt-image-1": "openai-gpt-image-1";
         "openai-gpt-image-1-mini": "openai-gpt-image-1-mini";
+        "bytedance-seedream-5-pro": "bytedance-seedream-5-pro";
+        "bytedance-seedream-5-flash": "bytedance-seedream-5-flash";
         "bytedance-seedream-5-lite": "bytedance-seedream-5-lite";
+        "bytedance-seedream-4.5": "bytedance-seedream-4.5";
+        "bytedance-seedream-4.0": "bytedance-seedream-4.0";
         "kling-kolors": "kling-kolors";
         "alibaba-qwen-image-2.0-pro": "alibaba-qwen-image-2.0-pro";
         "alibaba-qwen-image-2.0": "alibaba-qwen-image-2.0";
@@ -40,6 +42,8 @@ declare const InputSchema: z.ZodObject<{
         "minimax-image-01": "minimax-image-01";
         "runwayml-gen4-image": "runwayml-gen4-image";
         "runwayml-gen4-image-turbo": "runwayml-gen4-image-turbo";
+        "xai-grok-imagine-image-2.0": "xai-grok-imagine-image-2.0";
+        "xai-grok-imagine-image": "xai-grok-imagine-image";
     }>>;
     aspectRatio: z.ZodDefault<z.ZodEnum<{
         "9:16": "9:16";
@@ -56,8 +60,8 @@ declare const InputSchema: z.ZodObject<{
     seed: z.ZodOptional<z.ZodNumber>;
     numImages: z.ZodDefault<z.ZodNumber>;
     outputFormat: z.ZodOptional<z.ZodEnum<{
-        png: "png";
         jpeg: "jpeg";
+        png: "png";
         webp: "webp";
     }>>;
 }, z.core.$strip>;

package/lib/services/agent/tools/generateImage.tool.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"generateImage.tool.d.ts","sourceRoot":"","sources":["../../../../src/services/agent/tools/generateImage.tool.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AACH,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAKxB,OAAO,EAAgB,cAAc,EAAE,MAAM,iBAAiB,CAAC;AAwB/D,QAAA,MAAM,WAAW;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAkCf,CAAC;AAEH,QAAA,MAAM,YAAY;;;;;;;iBAOhB,CAAC;AA2BH,eAAO,MAAM,iBAAiB,EAAE,cAAc,CAC5C,OAAO,WAAW,EAClB,OAAO,YAAY,CAuGpB,CAAC"}
+{"version":3,"file":"generateImage.tool.d.ts","sourceRoot":"","sources":["../../../../src/services/agent/tools/generateImage.tool.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AACH,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAKxB,OAAO,EAAgB,cAAc,EAAE,MAAM,iBAAiB,CAAC;AAkC/D,QAAA,MAAM,WAAW;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAkCf,CAAC;AAEH,QAAA,MAAM,YAAY;;;;;;;iBAOhB,CAAC;AAkCH,eAAO,MAAM,iBAAiB,EAAE,cAAc,CAC5C,OAAO,WAAW,EAClB,OAAO,YAAY,CAuGpB,CAAC"}

package/lib/services/agent/tools/generateImage.tool.js

@@ -27,15 +27,23 @@ const rateLimiter_1 = require("../../rateLimiter");
 const toolRegistry_1 = require("../toolRegistry");
 const ModelKeyEnum = zod_1.z.enum([
     enums_1.EImageGenModels.GOOGLE_NANO_BANANA,
+    enums_1.EImageGenModels.GOOGLE_NANO_BANANA_2,
     enums_1.EImageGenModels.GOOGLE_NANO_BANANA_PRO,
-    enums_1.EImageGenModels.GOOGLE_IMAGEN_4_FAST,
-    enums_1.EImageGenModels.GOOGLE_IMAGEN_4,
-    enums_1.EImageGenModels.GOOGLE_IMAGEN_4_ULTRA,
+    // REMOVED FOR v2 — kept for reference: Imagen 4 keys are deprecated (Google
+    // shut the family down 2026-08-17; keys are repointed to Gemini 3.1 image
+    // models). Pick google-nano-banana / google-nano-banana-2 instead.
+    // EImageGenModels.GOOGLE_IMAGEN_4_FAST,
+    // EImageGenModels.GOOGLE_IMAGEN_4,
+    // EImageGenModels.GOOGLE_IMAGEN_4_ULTRA,
     enums_1.EImageGenModels.OPENAI_GPT_IMAGE_1_MINI,
     enums_1.EImageGenModels.OPENAI_GPT_IMAGE_1,
     enums_1.EImageGenModels.OPENAI_GPT_IMAGE_1_5,
     enums_1.EImageGenModels.OPENAI_GPT_IMAGE_2,
+    enums_1.EImageGenModels.BYTEDANCE_SEEDREAM_5_PRO,
+    enums_1.EImageGenModels.BYTEDANCE_SEEDREAM_5_FLASH,
     enums_1.EImageGenModels.BYTEDANCE_SEEDREAM_5_LITE,
+    enums_1.EImageGenModels.BYTEDANCE_SEEDREAM_4_5,
+    enums_1.EImageGenModels.BYTEDANCE_SEEDREAM_4_0,
     enums_1.EImageGenModels.KLING_KOLORS,
     enums_1.EImageGenModels.ALIBABA_QWEN_IMAGE_2,
     enums_1.EImageGenModels.ALIBABA_QWEN_IMAGE_2_PRO,
@@ -45,6 +53,8 @@ const ModelKeyEnum = zod_1.z.enum([
     enums_1.EImageGenModels.MINIMAX_IMAGE_01,
     enums_1.EImageGenModels.RUNWAYML_GEN4_IMAGE_TURBO,
     enums_1.EImageGenModels.RUNWAYML_GEN4_IMAGE,
+    enums_1.EImageGenModels.XAI_GROK_IMAGINE_IMAGE_2,
+    enums_1.EImageGenModels.XAI_GROK_IMAGINE_IMAGE,
 ]);
 const InputSchema = zod_1.z.object({
     prompt: zod_1.z
@@ -67,7 +77,7 @@ const InputSchema = zod_1.z.object({
         .array(zod_1.z.string().url())
         .max(16)
         .optional()
-        .describe("Multi-image references for identity transfer / composition. Used by AG-13 to thread character + product reference images into scene-level image gens. Per-provider caps: OpenAI ≤16, Nano Banana ≤14, Seedream ≤10, Qwen 1–3, MiniMax 1."),
+        .describe("Multi-image references for identity transfer / composition. Used by AG-13 to thread character + product reference images into scene-level image gens. Per-provider caps: OpenAI ≤16, Nano Banana ≤14, Seedream 5 pro/flash ≤10, other Seedream ≤14, Grok Imagine ≤5, Qwen 1–3, MiniMax 1."),
     negativePrompt: zod_1.z.string().optional(),
     seed: zod_1.z.number().int().optional(),
     numImages: zod_1.z.number().int().min(1).max(4).default(1),
@@ -86,15 +96,22 @@ const OutputSchema = zod_1.z.object({
 // budgeting in estimate_cost. Order roughly cheapest → most expensive.
 const COST_BY_MODEL = {
     [enums_1.EImageGenModels.GOOGLE_NANO_BANANA]: 1,
-    [enums_1.EImageGenModels.GOOGLE_IMAGEN_4_FAST]: 1,
+    [enums_1.EImageGenModels.GOOGLE_NANO_BANANA_2]: 1,
+    // [EImageGenModels.GOOGLE_IMAGEN_4_FAST]: 1, // REMOVED FOR v2 — deprecated
     [enums_1.EImageGenModels.OPENAI_GPT_IMAGE_1_MINI]: 1,
     [enums_1.EImageGenModels.GOOGLE_NANO_BANANA_PRO]: 2,
-    [enums_1.EImageGenModels.GOOGLE_IMAGEN_4]: 2,
+    // [EImageGenModels.GOOGLE_IMAGEN_4]: 2, // REMOVED FOR v2 — deprecated
+    [enums_1.EImageGenModels.BYTEDANCE_SEEDREAM_5_FLASH]: 1,
+    [enums_1.EImageGenModels.XAI_GROK_IMAGINE_IMAGE]: 1,
     [enums_1.EImageGenModels.BYTEDANCE_SEEDREAM_5_LITE]: 2,
+    [enums_1.EImageGenModels.BYTEDANCE_SEEDREAM_4_0]: 2,
+    [enums_1.EImageGenModels.BYTEDANCE_SEEDREAM_4_5]: 2,
+    [enums_1.EImageGenModels.XAI_GROK_IMAGINE_IMAGE_2]: 2,
     [enums_1.EImageGenModels.KLING_KOLORS]: 2,
     [enums_1.EImageGenModels.ALIBABA_QWEN_IMAGE_2]: 2,
     [enums_1.EImageGenModels.MINIMAX_IMAGE_01]: 2,
-    [enums_1.EImageGenModels.GOOGLE_IMAGEN_4_ULTRA]: 4,
+    // [EImageGenModels.GOOGLE_IMAGEN_4_ULTRA]: 4, // REMOVED FOR v2 — deprecated
+    [enums_1.EImageGenModels.BYTEDANCE_SEEDREAM_5_PRO]: 4,
     [enums_1.EImageGenModels.ALIBABA_QWEN_IMAGE_2_PRO]: 4,
     [enums_1.EImageGenModels.ALIBABA_QWEN_IMAGE_EDIT]: 4,
     [enums_1.EImageGenModels.ALIBABA_QWEN_IMAGE_EDIT_PLUS]: 4,

package/lib/services/agent/tools/generateVideo.tool.d.ts

@@ -19,6 +19,38 @@ declare const InputSchema: z.ZodObject<{
     lastFrameImageUrl: z.ZodOptional<z.ZodString>;
     audio: z.ZodOptional<z.ZodBoolean>;
     negativePrompt: z.ZodOptional<z.ZodString>;
+    cameraControl: z.ZodOptional<z.ZodObject<{
+        preset: z.ZodOptional<z.ZodEnum<{
+            static: "static";
+            pan_left: "pan_left";
+            pan_right: "pan_right";
+            tilt_up: "tilt_up";
+            tilt_down: "tilt_down";
+            zoom_in: "zoom_in";
+            zoom_out: "zoom_out";
+            dolly_in: "dolly_in";
+            dolly_out: "dolly_out";
+            orbit_left: "orbit_left";
+            orbit_right: "orbit_right";
+            roll_left: "roll_left";
+            roll_right: "roll_right";
+            tracking: "tracking";
+            handheld: "handheld";
+            push_in: "push_in";
+            pull_out: "pull_out";
+        }>>;
+        vector: z.ZodOptional<z.ZodObject<{
+            horizontal: z.ZodOptional<z.ZodNumber>;
+            vertical: z.ZodOptional<z.ZodNumber>;
+            pan: z.ZodOptional<z.ZodNumber>;
+            tilt: z.ZodOptional<z.ZodNumber>;
+            roll: z.ZodOptional<z.ZodNumber>;
+            zoom: z.ZodOptional<z.ZodNumber>;
+        }, z.core.$strip>>;
+        intensity: z.ZodOptional<z.ZodNumber>;
+    }, z.core.$strip>>;
+    referenceImageUrls: z.ZodOptional<z.ZodArray<z.ZodString>>;
+    seed: z.ZodOptional<z.ZodNumber>;
 }, z.core.$strip>;
 declare const OutputSchema: z.ZodObject<{
     status: z.ZodLiteral<"triggered">;

package/lib/services/agent/tools/generateVideo.tool.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"generateVideo.tool.d.ts","sourceRoot":"","sources":["../../../../src/services/agent/tools/generateVideo.tool.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,eAAe,EAAE,MAAM,2BAA2B,CAAC;AAI5D,OAAO,EAAgB,cAAc,EAAE,MAAM,iBAAiB,CAAC;AA6B/D,QAAA,MAAM,WAAW;;;;;;;;;;;;;;;;;;iBA4Cf,CAAC;AAEH,QAAA,MAAM,YAAY;;;;;;iBAOhB,CAAC;AAEH,eAAO,MAAM,iBAAiB,EAAE,cAAc,CAC5C,OAAO,WAAW,EAClB,OAAO,YAAY,CAqHpB,CAAC"}
+{"version":3,"file":"generateVideo.tool.d.ts","sourceRoot":"","sources":["../../../../src/services/agent/tools/generateVideo.tool.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,eAAe,EAAE,MAAM,2BAA2B,CAAC;AAM5D,OAAO,EAAgB,cAAc,EAAE,MAAM,iBAAiB,CAAC;AAgC/D,QAAA,MAAM,WAAW;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAyDf,CAAC;AAEH,QAAA,MAAM,YAAY;;;;;;iBAOhB,CAAC;AAEH,eAAO,MAAM,iBAAiB,EAAE,cAAc,CAC5C,OAAO,WAAW,EAClB,OAAO,YAAY,CA0IpB,CAAC"}