@shop-os/modelkit 0.7.8 → 0.7.11

package/dist/index.js

@@ -4706,6 +4706,8 @@ function range(min, max) {
 }
 var KLING_PROMPT = { referenceSyntax: "numbered", maxChars: 2500 };
 var H3_PROMPT = { referenceSyntax: "numbered", maxChars: 5e4 };
+var KLING_V4_PROMPT = { referenceSyntax: "numbered", maxChars: 2500 };
+var KLING_V4_AUDIO = "always";
 var NO_REFERENCES = { images: 0, videos: 0, audios: 0 };
 function klingMotionControl(tier) {
   const label = tier === "pro" ? "Pro" : "Standard";
@@ -4806,6 +4808,79 @@ var VIDEO_FAMILY_MODELS = [
     },
     capabilities: { acceptsSourceVideo: true, videoEdit: true }
   },
+  {
+    // https://fal.ai/models/kling/v4/reference-to-video/api (+ /text-to-video,
+    // /image-to-video): one row; references (images or videos) go to reference-to-video,
+    // else a start frame to image-to-video, else text-to-video. Reference images: 10, the
+    // start frame included (fal: "counts toward the image reference limit"), 300 px min.
+    // Reference videos: 5, 3-30.05 s each and 30 s combined, 700 px min, 23.9-60.05 fps.
+    // The prompt names references as @Image1, @Video1 (no space; @Element1 for elements,
+    // which the kit does not send). No sound switch on any endpoint.
+    id: "kling-4.0",
+    aliases: ["kling-4", "kling-v4"],
+    label: "Kling 4.0",
+    type: "video",
+    adapter: "fal",
+    providerModelId: "kling/v4",
+    lifecycle: "active",
+    exposure: ["cowork", "space_builder"],
+    constraints: {
+      aspectRatios: ["adaptive", "16:9", "9:16", "1:1", "21:9"],
+      resolutions: ["720p", "1080p"],
+      durations: { min: 3, max: 30 },
+      durationPresets: [5, 10, 15, 30],
+      references: {
+        ...NO_REFERENCES,
+        images: 10,
+        videos: 5,
+        videoSeconds: { min: 3, max: 30.05, total: 30 }
+      },
+      supports: ["lastFrame"],
+      prompt: KLING_V4_PROMPT,
+      audio: KLING_V4_AUDIO
+    },
+    defaults: { durationSeconds: 5, resolution: "1080p", aspectRatio: "16:9" },
+    usage: { ...VIDEO_USAGE },
+    pricingKind: "video_per_second",
+    display: {
+      provider: "kuaishou",
+      logo: KLING_LOGO,
+      description: "Kuaishou Kling 4.0: text-to-video, start/end frames, or reference-to-video from up to 10 images and 5 videos (30 seconds combined), 3-30 seconds at 720p or 1080p. Prompts name references @Image1, @Video1."
+    },
+    capabilities: { lastFrameRequiresFirstFrame: true }
+  },
+  {
+    // https://fal.ai/models/kling/v4/video-to-video/edit/api: source 3-30.05 s (700 px
+    // min, 23.9-60.05 fps), up to 10 reference images. duration 3-30, or left out for the
+    // source's length. The prompt names the source @Video1 and images @Image1, @Image2.
+    id: "kling-4.0-edit",
+    aliases: ["kling-4-edit", "kling-v4-edit"],
+    label: "Kling 4.0 Edit",
+    type: "video",
+    adapter: "fal",
+    providerModelId: "kling/v4/video-to-video/edit",
+    lifecycle: "active",
+    exposure: ["cowork", "space_builder"],
+    constraints: {
+      resolutions: ["720p", "1080p"],
+      durations: { min: 3, max: 30 },
+      references: { ...NO_REFERENCES, images: 10 },
+      requires: ["sourceVideo"],
+      sourceVideoSeconds: { min: 3, max: 30.05 },
+      supports: [],
+      prompt: KLING_V4_PROMPT,
+      audio: KLING_V4_AUDIO
+    },
+    defaults: { resolution: "1080p" },
+    usage: { ...VIDEO_USAGE },
+    pricingKind: "video_per_second",
+    display: {
+      provider: "kuaishou",
+      logo: KLING_LOGO,
+      description: "Kuaishou Kling 4.0 video edit: change a 3-30 second clip with a prompt and up to 10 reference images, at 720p or 1080p, as long as the source or a set 3-30 seconds. Prompts name the source @Video1 and images @Image1."
+    },
+    capabilities: { acceptsSourceVideo: true, videoEdit: true }
+  },
   {
     // https://fal.ai/models/minimax/h3-max-turbo/text-to-video/api (+ /image-to-video).
     // image-to-video takes a start frame, an end frame, or both, and no aspect_ratio.
@@ -5664,38 +5739,38 @@ var VIDEO_MODELS = [
     capabilities: { ...FRAMES_EXCLUDE_REFERENCES }
   },
   {
-    // mode standard/pro/4k selects 720p/1080p/4K; the endpoint takes no separate resolution.
-    // Replicate: "Max 7 [images] without video, 4 with video"; 4K does not support
-    // reference_video. No `videos` limit yet: the adapter sends no reference video,
-    // so `imagesWithVideo` stays dormant until one is wired through.
+    // fal Kling O3 (adapters/fal-kling-omni.ts): mode standard/pro/4k selects 720p/1080p/4K
+    // in the endpoint path. Moved off Replicate kwaivgi/kling-v3-omni-video, which fal sells
+    // at half the price; the id and inputs stay. The Replicate mapper stays for deployed
+    // model_catalog rows not yet switched. Frames and reference images share fal's cap of 7.
+    // Reference video, multi-shot and keep-sound were never wired, so stay off.
     id: "kling-v3-omni-video",
     aliases: ["kwaivgi/kling-v3-omni-video", "kling-v3-omni"],
     label: "Kling V3 Omni",
     type: "video",
-    adapter: "replicate",
-    providerModelId: "kwaivgi/kling-v3-omni-video",
+    adapter: "fal",
+    providerModelId: "fal-ai/kling-video/o3",
     lifecycle: "active",
     exposure: ["cowork", "space_builder"],
     constraints: {
       aspectRatios: [...KLING_OMNI_RATIOS],
       resolutions: ["720p", "1080p", "4k"],
       durations: { min: 3, max: 15 },
-      references: { images: 7, imagesWithVideo: 4 },
+      references: { images: 7, videos: 0, audios: 0 },
       supports: ["generateAudio", "lastFrame"],
       prompt: KLING_PROMPT2,
       audio: "optional"
     },
     defaults: { durationSeconds: 5, aspectRatio: "16:9", generateAudio: true, mode: "pro" },
-    usage: { unit: "compute_seconds" },
+    usage: { ...VIDEO_USAGE3 },
     pricingKind: "video_per_second",
     display: {
       provider: "kuaishou",
       logo: "https://cdn.jsdelivr.net/npm/@lobehub/icons-static-png@1.97.0/light/kling-color.png",
-      description: "Kuaishou Kling Video 3.0 Omni unified generation and editing from text, start/end images, up to 7 reference images, and reference video; up to 15 seconds at up to 4K."
+      description: "Kuaishou Kling Video 3.0 Omni generation from text, start/end images and up to 7 reference images, with optional native audio; up to 15 seconds at up to 4K."
     }
-    // end_image "Requires start_image", but no lastFrameRequiresFirstFrame: SpaceLab resolves
-    // replicate rows in error mode, so the flag would refuse its end-frame-only nodes, which
-    // the adapter sends as start_image.
+    // No lastFrameRequiresFirstFrame: fal image-to-video needs a start frame, so the adapter
+    // sends a lone end frame as the start frame, as the Replicate model took it.
   },
   {
     id: "minimax-h3",
@@ -9310,11 +9385,11 @@ function toolInput(ctx, req, output) {
 }
 var EXR_DEFAULT_PROMPT = "The same video in high dynamic range, with unchanged content, motion, framing and timing.";
 async function submitFalVideoTool(req, ctx, opts) {
-  const plan3 = planVideoTool(ctx.model, req);
-  if (!plan3.ok) {
+  const plan4 = planVideoTool(ctx.model, req);
+  if (!plan4.ok) {
     throw new ModelKitError({
       code: "unsupported_option",
-      message: plan3.error,
+      message: plan4.error,
       provider: "fal",
       model: ctx.model.id,
       retryable: false
@@ -9322,7 +9397,7 @@ async function submitFalVideoTool(req, ctx, opts) {
   }
   const target = new URL(`${FAL_QUEUE_ORIGIN}/${ctx.model.providerModelId}`);
   if (opts.webhookUrl !== void 0) target.searchParams.set("fal_webhook", opts.webhookUrl);
-  const res = await queueFetch(ctx, target.toString(), "POST", toolInput(ctx, req, plan3.output));
+  const res = await queueFetch(ctx, target.toString(), "POST", toolInput(ctx, req, plan4.output));
   const parsed = submitResponseSchema.safeParse(await readJson(ctx, res));
   if (!parsed.success) {
     throw new ModelKitError({
@@ -9340,7 +9415,7 @@ async function submitFalVideoTool(req, ctx, opts) {
     statusUrl: parsed.data.status_url,
     responseUrl: parsed.data.response_url,
     cancelUrl: parsed.data.cancel_url,
-    output: plan3.output
+    output: plan4.output
   };
 }
 async function falVideoToolStatus(sub, ctx) {
@@ -10029,14 +10104,14 @@ function planFalImage(req, providerModelId, model) {
   const exact = falExactSize(spec, req);
   const call = { label: spec.label, req, refs, edit: edit2, ...exact !== void 0 && { exact } };
   const built = family.build(call, model);
-  const plan3 = {
+  const plan4 = {
     endpoint: edit2 && family.editEndpoint !== void 0 ? family.editEndpoint : id,
     input: built.input,
     output: family.output ?? "images"
   };
-  if (built.imageQuality !== void 0) plan3.imageQuality = built.imageQuality;
-  if (built.imageSize !== void 0) plan3.imageSize = built.imageSize;
-  return plan3;
+  if (built.imageQuality !== void 0) plan4.imageQuality = built.imageQuality;
+  if (built.imageSize !== void 0) plan4.imageSize = built.imageSize;
+  return plan4;
 }
 var FalFileSchema = z14.object({ url: z14.string(), content_type: z14.string().nullish() });
 var ImagesPayloadSchema = z14.object({ images: z14.array(FalFileSchema) });
@@ -10058,18 +10133,18 @@ async function downloadFalFile(ctx, file) {
   const mime = res.headers.get("content-type") ?? file.content_type ?? "";
   return { bytes: new Uint8Array(await res.arrayBuffer()), mime };
 }
-function filesOf(payload, plan3) {
-  if (plan3.output === "image") {
+function filesOf(payload, plan4) {
+  if (plan4.output === "image") {
     const parsed2 = ImagePayloadSchema.safeParse(payload);
     return parsed2.success ? [parsed2.data.image] : [];
   }
   const parsed = ImagesPayloadSchema.safeParse(payload);
   return parsed.success ? parsed.data.images : [];
 }
-async function collectFalPlanImages(payload, plan3, ctx, req) {
-  const files = filesOf(payload, plan3);
+async function collectFalPlanImages(payload, plan4, ctx, req) {
+  const files = filesOf(payload, plan4);
   if (files.length === 0) {
-    throw fail2(ctx.model.id, `fal ${plan3.endpoint} returned no image`, "provider_rejected");
+    throw fail2(ctx.model.id, `fal ${plan4.endpoint} returned no image`, "provider_rejected");
   }
   const images = await Promise.all(
     files.map(async (file) => {
@@ -10084,18 +10159,18 @@ async function collectFalPlanImages(payload, plan3, ctx, req) {
     unit: "images",
     quantity: images.length
   };
-  const billedSize = plan3.imageSize ?? req.size;
+  const billedSize = plan4.imageSize ?? req.size;
   if (billedSize !== void 0 && ctx.model.type === "image") usage3.imageSize = billedSize;
-  if (plan3.imageQuality !== void 0) usage3.imageQuality = plan3.imageQuality;
+  if (plan4.imageQuality !== void 0) usage3.imageQuality = plan4.imageQuality;
   const result = { images, usage: usage3, raw: payload };
   const remote = files.find((f) => !f.url.startsWith("data:"));
   if (remote !== void 0) result.outputUrl = remote.url;
   return result;
 }
 async function generateFalFamilyImage(req, ctx, run) {
-  const plan3 = planFalImage(req, ctx.model.providerModelId, ctx.model.id);
-  const payload = await run(plan3.endpoint, plan3.input);
-  return collectFalPlanImages(payload, plan3, ctx, req);
+  const plan4 = planFalImage(req, ctx.model.providerModelId, ctx.model.id);
+  const payload = await run(plan4.endpoint, plan4.input);
+  return collectFalPlanImages(payload, plan4, ctx, req);
 }
 
 // src/adapters/fal-mask.ts
@@ -11113,6 +11188,8 @@ var MOTION_PRO = "fal-ai/kling-video/v3/pro/motion-control";
 var MOTION_STANDARD = "fal-ai/kling-video/v3/standard/motion-control";
 var O3_REFERENCE = "fal-ai/kling-video/o3/pro/reference-to-video";
 var O3_EDIT = "fal-ai/kling-video/o3/pro/video-to-video/edit";
+var V4_FAMILY = "kling/v4";
+var V4_EDIT = "kling/v4/video-to-video/edit";
 function motionControl(endpoint, req, ctx) {
   const [character] = req.referenceImages ?? [];
   const [motion] = req.referenceVideos ?? [];
@@ -11172,6 +11249,61 @@ var klingV3Family = {
     return motionControl(id, req, ctx);
   }
 };
+function v4Generate(req, ctx) {
+  const images = urls(req.referenceImages);
+  const videos = urls(req.referenceVideos);
+  const duration = wholeSeconds(req, ctx);
+  const resolution = resolutionOf(req, ctx);
+  const input = { prompt: req.prompt.trim(), duration, resolution };
+  let endpoint;
+  if (images.length + videos.length > 0) {
+    endpoint = `${V4_FAMILY}/reference-to-video`;
+    if (images.length > 0) input["image_urls"] = images;
+    if (videos.length > 0) input["video_urls"] = videos;
+    if (req.firstFrame) input["start_image_url"] = refToUrl(req.firstFrame);
+    if (req.aspectRatio !== void 0) {
+      input["aspect_ratio"] = req.aspectRatio === "adaptive" ? "auto" : req.aspectRatio;
+    }
+  } else if (req.firstFrame) {
+    endpoint = `${V4_FAMILY}/image-to-video`;
+    input["start_image_url"] = refToUrl(req.firstFrame);
+    if (req.lastFrame) input["end_image_url"] = refToUrl(req.lastFrame);
+  } else {
+    endpoint = `${V4_FAMILY}/text-to-video`;
+    const ratio = concreteRatio(req);
+    if (ratio !== void 0) input["aspect_ratio"] = ratio;
+  }
+  return { endpoint, input, resolution, seconds: requestedSeconds(duration) };
+}
+function v4Edit(req, ctx) {
+  if (req.sourceVideo === void 0) throw refusal(ctx, "needs a source video", "missing_input");
+  const resolution = resolutionOf(req, ctx);
+  const input = {
+    prompt: req.prompt.trim(),
+    video_url: refToUrl(req.sourceVideo),
+    resolution
+  };
+  const images = urls(req.referenceImages);
+  if (images.length > 0) input["image_urls"] = images;
+  if (req.durationSeconds === void 0) {
+    return {
+      endpoint: V4_EDIT,
+      input,
+      resolution,
+      seconds: inputLengthSeconds(req, req.sourceVideo)
+    };
+  }
+  const duration = wholeSeconds(req, ctx);
+  input["duration"] = duration;
+  return { endpoint: V4_EDIT, input, resolution, seconds: requestedSeconds(duration) };
+}
+var KLING_V4_VIDEO_IDS = /* @__PURE__ */ new Set([V4_FAMILY, V4_EDIT]);
+var klingV4Family = {
+  ids: KLING_V4_VIDEO_IDS,
+  plan(req, ctx) {
+    return ctx.model.providerModelId.toLowerCase() === V4_EDIT ? v4Edit(req, ctx) : v4Generate(req, ctx);
+  }
+};
 
 // src/replicate-video-kinds.ts
 function replicateBaseModelId(providerModelId) {
@@ -11193,15 +11325,57 @@ var KLING_TIERS = [
   { mode: "pro", resolution: "1080p" },
   { mode: "4k", resolution: "4k" }
 ];
+var KLING_O3_OMNI = "fal-ai/kling-video/o3";
 function klingTier(req) {
   return KLING_TIERS.find((tier) => tier.resolution === req.resolution) ?? KLING_TIERS.find((tier) => tier.mode === req.mode) ?? KLING_TIERS[0];
 }
 
+// src/adapters/fal-kling-omni.ts
+function plan(req, ctx) {
+  const tier = klingTier(req);
+  const duration = wholeSeconds(req, ctx);
+  const input = {
+    prompt: req.prompt.trim(),
+    duration: String(duration),
+    generate_audio: req.generateAudio ?? true
+  };
+  const start = req.firstFrame ?? req.lastFrame;
+  const end = req.firstFrame ? req.lastFrame : void 0;
+  const images = urls(req.referenceImages);
+  let task;
+  if (images.length > 0) {
+    task = "reference-to-video";
+    input["image_urls"] = images;
+    if (start) input["start_image_url"] = refToUrl(start);
+    if (end) input["end_image_url"] = refToUrl(end);
+  } else if (start) {
+    task = "image-to-video";
+    input["image_url"] = refToUrl(start);
+    if (end) input["end_image_url"] = refToUrl(end);
+  } else {
+    task = "text-to-video";
+  }
+  const ratio = concreteRatio(req);
+  if (ratio !== void 0 && start === void 0) input["aspect_ratio"] = ratio;
+  return {
+    endpoint: `${KLING_O3_OMNI}/${tier.mode}/${task}`,
+    input,
+    resolution: tier.resolution,
+    seconds: requestedSeconds(duration)
+  };
+}
+var klingOmniFamily = {
+  ids: /* @__PURE__ */ new Set([KLING_O3_OMNI]),
+  plan
+};
+
 // src/resolve-video-rules.ts
 var MOTION_PRO2 = "fal-ai/kling-video/v3/pro/motion-control";
 var MOTION_STANDARD2 = "fal-ai/kling-video/v3/standard/motion-control";
 var O3_REFERENCE2 = "fal-ai/kling-video/o3/pro/reference-to-video";
 var O3_EDIT2 = "fal-ai/kling-video/o3/pro/video-to-video/edit";
+var KLING_V4 = "kling/v4";
+var KLING_V4_EDIT = "kling/v4/video-to-video/edit";
 var H3_TURBO = "minimax/h3-max-turbo";
 var H3_RECAST = "minimax/h3-max/recast";
 var H3_RELIGHT = "minimax/h3-max/relight";
@@ -11210,6 +11384,7 @@ var HAPPY_HORSE_EDIT = "alibaba/happy-horse/video-edit";
 var VIDU_Q4 = "fal-ai/vidu/q4";
 var OMNIHUMAN_V2 = "bytedance/omnihuman/v2";
 var KLING_O3_IMAGE_CAP = 7;
+var KLING_V4_IMAGE_CAP = 10;
 var TURBO_AUDIO_MIN_SECONDS = 2;
 var OMNIHUMAN_AUDIO_SECONDS = { min: 4, max: 300 };
 function missing(message) {
@@ -11265,6 +11440,23 @@ function o3Reference2(req) {
   }
   return needsPrompt(req);
 }
+function hasReferences(req) {
+  return count(req.referenceImages) + count(req.referenceVideos) > 0;
+}
+function klingV4(req) {
+  if (hasReferences(req)) {
+    if (req.lastFrame !== void 0) {
+      return unsupported2("takes an end frame only without reference images or videos");
+    }
+    const images = count(req.referenceImages) + (req.firstFrame === void 0 ? 0 : 1);
+    if (images > KLING_V4_IMAGE_CAP) {
+      return unsupported2(
+        `takes up to ${KLING_V4_IMAGE_CAP} images in all, the start frame included; this request has ${images}`
+      );
+    }
+  }
+  return needsPrompt(req);
+}
 function h3Turbo(req) {
   if (count(req.referenceImages) + count(req.referenceVideos) > 0) {
     return unsupported2("takes no reference images or videos");
@@ -11341,7 +11533,10 @@ var RULES = /* @__PURE__ */ new Map([
   [MOTION_PRO2, motionControl2],
   [MOTION_STANDARD2, motionControl2],
   [O3_REFERENCE2, o3Reference2],
+  [KLING_O3_OMNI, o3Reference2],
   [O3_EDIT2, (req) => needsSource(req) ?? needsPrompt(req)],
+  [KLING_V4, klingV4],
+  [KLING_V4_EDIT, (req) => needsSource(req) ?? needsPrompt(req)],
   [H3_TURBO, h3Turbo],
   [MINIMAX_H3_INSERT, h3Insert],
   [H3_RECAST, h3Recast],
@@ -11370,13 +11565,14 @@ function h3EndFrameRatioWarning(model, ratio) {
     message: `${model.id} follows the end frame's aspect ratio`
   };
 }
-function isReplicateKling(model) {
+function takesKlingMode(model) {
+  if (model.adapter === "fal") return model.providerModelId.toLowerCase() === KLING_O3_OMNI;
   if (model.adapter !== "replicate") return false;
   const kind = videoKind(model.providerModelId);
   return kind === "kling" || kind === "kling-omni";
 }
-function fillReplicateKlingResolution(model, req) {
-  if (isReplicateKling(model) && req.resolution === void 0) {
+function fillKlingModeResolution(model, req) {
+  if (takesKlingMode(model) && req.resolution === void 0) {
     req.resolution = klingTier(req).resolution;
   }
 }
@@ -11390,6 +11586,19 @@ function applyFalVideoRules(model, req, callerRatio, warnings) {
     warnings.push(viduSilentWarning(model));
     req.generateAudio = false;
   }
+  if (id === KLING_V4 && req.firstFrame !== void 0 && !hasReferences(req)) {
+    if (req.aspectRatio !== void 0) {
+      if (callerRatio !== void 0) {
+        warnings.push({
+          code: "ratio_dropped",
+          field: "aspectRatio",
+          from: req.aspectRatio,
+          message: `${model.id} follows the start frame's aspect ratio`
+        });
+      }
+      delete req.aspectRatio;
+    }
+  }
   const endFrameOnly = req.lastFrame !== void 0 && req.firstFrame === void 0;
   if (id === H3_TURBO && endFrameOnly && req.aspectRatio !== void 0) {
     if (callerRatio !== void 0 && callerRatio !== "adaptive") {
@@ -11515,7 +11724,7 @@ var minimaxH3MaxFamily = {
 
 // src/adapters/fal-omnihuman.ts
 var OMNIHUMAN_V22 = "bytedance/omnihuman/v2";
-function plan(req, ctx) {
+function plan2(req, ctx) {
   const [image] = req.referenceImages ?? [];
   const [audio] = req.referenceAudios ?? [];
   if (image === void 0 || audio === void 0) {
@@ -11526,14 +11735,14 @@ function plan(req, ctx) {
   return { endpoint: OMNIHUMAN_V22, input, seconds: inputLengthSeconds(req, audio) };
 }
 var OMNIHUMAN_MODEL_IDS = /* @__PURE__ */ new Set([OMNIHUMAN_V22]);
-var omnihumanFamily = { ids: OMNIHUMAN_MODEL_IDS, plan };
+var omnihumanFamily = { ids: OMNIHUMAN_MODEL_IDS, plan: plan2 };
 
 // src/adapters/fal-vidu.ts
 var FAMILY2 = "fal-ai/vidu/q4";
 function viduTier(resolution) {
   return resolution === "2k" || resolution === "4k" ? upperTier(resolution) : resolution;
 }
-function plan2(req, ctx) {
+function plan3(req, ctx) {
   const images = urls(req.referenceImages);
   const audios = urls(req.referenceAudios);
   const duration = wholeSeconds(req, ctx);
@@ -11560,11 +11769,13 @@ function plan2(req, ctx) {
   return { endpoint, input, resolution, seconds: requestedSeconds(duration) };
 }
 var VIDU_MODEL_IDS = /* @__PURE__ */ new Set([FAMILY2]);
-var viduFamily = { ids: VIDU_MODEL_IDS, plan: plan2 };
+var viduFamily = { ids: VIDU_MODEL_IDS, plan: plan3 };
 
 // src/adapters/fal-video-families.ts
 var FAMILIES2 = [
   klingV3Family,
+  klingOmniFamily,
+  klingV4Family,
   minimaxH3MaxFamily,
   happyHorseFamily,
   viduFamily,
@@ -11851,8 +12062,8 @@ async function collectFalImages(payload, ctx, req, label) {
   };
 }
 async function generateSpecImage(req, ctx, deps, spec) {
-  const hasReferences = (req.references ?? []).length > 0;
-  const endpoint = hasReferences && spec.spec.editEndpoint !== void 0 ? spec.spec.editEndpoint : ctx.model.providerModelId;
+  const hasReferences2 = (req.references ?? []).length > 0;
+  const endpoint = hasReferences2 && spec.spec.editEndpoint !== void 0 ? spec.spec.editEndpoint : ctx.model.providerModelId;
   const payload = await runJob(ctx, deps, endpoint, spec.input);
   return collectFalImages(payload, ctx, req, `fal ${endpoint}`);
 }
@@ -12041,7 +12252,7 @@ function extractVideoUrl(result, endpoint, model) {
 }
 async function generateFalVideo(req, ctx, deps) {
   const providerModelId = ctx.model.providerModelId;
-  const plan3 = planFalFamilyVideo(req, ctx);
+  const plan4 = planFalFamilyVideo(req, ctx);
   const start = (endpoint2, input, usageOf2) => startFalVideoJob(
     ctx,
     () => falClientOf(ctx, deps),
@@ -12049,7 +12260,7 @@ async function generateFalVideo(req, ctx, deps) {
     input,
     videoFileDecoder(ctx, req, endpoint2, usageOf2)
   );
-  if (plan3 !== void 0) return start(plan3.endpoint, plan3.input, planUsage(ctx, plan3));
+  if (plan4 !== void 0) return start(plan4.endpoint, plan4.input, planUsage(ctx, plan4));
   if (isKling(providerModelId)) return start(KLING_O1_ENDPOINT, klingO1Input(req));
   if (isWan(providerModelId)) {
     const { endpoint: endpoint2, input } = wanVideo(req, providerModelId.toLowerCase());
@@ -12099,9 +12310,9 @@ function videoFileDecoder(ctx, req, endpoint, usageOf2) {
     raw: payload
   });
 }
-function planUsage(ctx, plan3) {
+function planUsage(ctx, plan4) {
   return (payload) => {
-    const seconds3 = plan3.seconds(payload);
+    const seconds3 = plan4.seconds(payload);
     return {
       model: ctx.model.id,
       provider: "fal",
@@ -12109,7 +12320,7 @@ function planUsage(ctx, plan3) {
       unit: "seconds",
       quantity: seconds3,
       durationSeconds: seconds3,
-      ...plan3.resolution !== void 0 && { resolution: plan3.resolution }
+      ...plan4.resolution !== void 0 && { resolution: plan4.resolution }
     };
   };
 }
@@ -14199,8 +14410,8 @@ function imageBody(req, ctx, outputFormat) {
 function videoBody(req, ctx, mode) {
   const prompt = checkPrompt(req, ctx);
   const model = lumaModel(ctx);
-  const hasReferences = (req.referenceImages ?? []).length + (req.referenceVideos ?? []).length + (req.referenceAudios ?? []).length > 0;
-  if (hasReferences) {
+  const hasReferences2 = (req.referenceImages ?? []).length + (req.referenceVideos ?? []).length + (req.referenceAudios ?? []).length > 0;
+  if (hasReferences2) {
     refuseOrDrop(
       ctx,
       "referenceImages",
@@ -16563,10 +16774,10 @@ function resolveVideoOptions(model, request, onUnsupported = "error") {
     warnings
   );
   if (aspectRatio !== void 0) merged.aspectRatio = aspectRatio;
-  fillReplicateKlingResolution(model, merged);
+  fillKlingModeResolution(model, merged);
   const resolution = resolveVideoResolution(merged.resolution, model, onUnsupported, warnings);
   if (resolution !== void 0) merged.resolution = resolution;
-  if (isReplicateKling(model)) delete merged.mode;
+  if (takesKlingMode(model)) delete merged.mode;
   const duration = resolveDuration(
     merged.durationSeconds,
     c.durations,