@shop-os/modelkit 0.7.10 → 0.8.0

package/CHANGELOG.md

@@ -9,6 +9,24 @@ header in the release PR.
 
 ## [Unreleased]
 
+## [0.8.0] - Unreleased
+
+The BytePlus ModelArk key has one env name.
+
+### Removed
+- `keysFromEnv` reads the `ark` key from `BYTEDANCE_API_KEY` only (with the `ENTERPRISE_` / `SLOOSH_` prefixes); the two older alias names are no longer read. An env that sets only an alias gets no `ark` key, so an Ark call fails with a `ModelKitError` code `auth`. Set `BYTEDANCE_API_KEY` before a consumer upgrades.
+
+## [0.7.11] - Unreleased
+
+Kling 4.0 on fal.
+
+### Added
+- `kling-4.0` (fal `kling/v4`): one row for text-, image- and reference-to-video. Reference images or videos go to `kling/v4/reference-to-video` (`image_urls`, `video_urls`, and a start frame as `start_image_url`), else a start frame to `image-to-video` (with an optional end frame), else `text-to-video`. 720p / 1080p, 3-30 s (whole seconds), aspect ratios `adaptive` (fal's `auto` on reference-to-video), 16:9, 9:16, 1:1, 21:9. References: 10 images (the start frame counts toward them) and 5 videos, each 3-30.05 s and 30 s combined. An end frame with references, more than 10 images with the start frame, and a blank prompt are refused before the call.
+- `kling-4.0-edit` (fal `kling/v4/video-to-video/edit`): a 3-30.05 s source plus up to 10 reference images, 720p / 1080p. `durationSeconds` 3-30 sets the output length; left out, the output (and its usage) is the source's length.
+- Prompts name inputs `@Image1`, `@Video1` (no space); on an edit the source is `@Video1`. Kit requests send no Kling elements.
+- Neither endpoint has a sound field, so both rows declare `audio: "always"` and drop `generateAudio` with an `option_ignored` warning.
+- Usage reports the output seconds and resolution. fal bills a request with reference videos, and every edit, on input plus output seconds (at least twice the output) at a lower rate; pricing that is the consumer's job (pro-backend's video-input rate).
+
 ## [0.7.10] - Unreleased
 
 Kling V3 Omni moves from Replicate to fal at half the provider cost.

package/README.md

@@ -108,7 +108,7 @@ before submit), and `model3dPricingSku(row, options, inputs)` names the flat pri
 bills under; `model3dSkus(row)` lists them all, and pro-backend keeps a `credit_pricing_flat`
 row for each. All three are in the browser-safe catalog entry.
 
-Keys come from `process.env` by default: bare names for pro (`OPENAI_API_KEY`, `FAL_KEY`, `ARK_API_KEY`, `LUMA_API_KEY` ...), the `ENTERPRISE_` prefix for enterprise and the `SLOOSH_` prefix for sloosh (sloosh reads `SLOOSH_LUMA_API_KEY` for Luma). Pass `keys: { pro, enterprise, sloosh }` to override. `tier: "sloosh"` uses the sloosh keys only: a missing key is a `ModelKitError` with code `auth`, never a fallback to pro.
+Keys come from `process.env` by default: bare names for pro (`OPENAI_API_KEY`, `FAL_KEY`, `BYTEDANCE_API_KEY`, `LUMA_API_KEY` ...), the `ENTERPRISE_` prefix for enterprise and the `SLOOSH_` prefix for sloosh (sloosh reads `SLOOSH_LUMA_API_KEY` for Luma). Pass `keys: { pro, enterprise, sloosh }` to override. `tier: "sloosh"` uses the sloosh keys only: a missing key is a `ModelKitError` with code `auth`, never a fallback to pro.
 
 ## Browser-safe catalog
 

package/dist/catalog.cjs

@@ -4778,6 +4778,8 @@ 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";
@@ -4878,6 +4880,79 @@ 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.
@@ -8532,6 +8607,8 @@ 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 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";
@@ -8540,6 +8617,7 @@ 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) {
@@ -8595,6 +8673,23 @@ function o3Reference(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 ${images}`
+      );
+    }
+  }
+  return needsPrompt(req);
+}
 function h3Turbo(req) {
   if (count(req.referenceImages) + count(req.referenceVideos) > 0) {
     return unsupported2("takes no reference images or videos");
@@ -8673,6 +8768,8 @@ var RULES = /* @__PURE__ */ new Map([
   [O3_REFERENCE, o3Reference],
   [KLING_O3_OMNI, o3Reference],
   [O3_EDIT, (req) => needsSource(req) ?? needsPrompt(req)],
+  [KLING_V4, klingV4],
+  [KLING_V4_EDIT, (req) => needsSource(req) ?? needsPrompt(req)],
   [H3_TURBO, h3Turbo],
   [MINIMAX_H3_INSERT, h3Insert],
   [H3_RECAST, h3Recast],
@@ -8722,6 +8819,19 @@ function applyFalVideoRules(model, req, callerRatio, warnings) {
     warnings.push(viduSilentWarning(model));
     req.generateAudio = false;
   }
+  if (id === KLING_V4 && req.firstFrame !== void 0 && !hasReferences(req)) {
+    if (req.aspectRatio !== void 0) {
+      if (callerRatio !== void 0) {
+        warnings.push({
+          code: "ratio_dropped",
+          field: "aspectRatio",
+          from: req.aspectRatio,
+          message: `${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") {