@koda-sl/baker-cli 0.246.0 → 0.247.0

package/README.md

@@ -3485,7 +3485,7 @@ baker studio animate "handheld drift right, steam rising from the cup" \
 baker studio animate "product rotates once on a turntable" --image j57abc…,j57def… --from references
 ```
 
-The image fixes the look, so the prompt describes **movement**. Models: `google/gemini-omni-flash` (default), `bytedance/seedance-2.0`, `bytedance/seedance-2.5`, `google/veo-3.1`, `google/veo-3.1-fast`, `kwaivgi/kling-v3.0-pro`.
+The image fixes the look, so the prompt describes **movement**. Models: `google/gemini-omni-1.1-flash` (default), `google/gemini-omni-flash` (legacy preview), `bytedance/seedance-2.0`, `bytedance/seedance-2.5`, `google/veo-3.1`, `google/veo-3.1-fast`, `kwaivgi/kling-v3.0-pro`.
 
 **`bytedance/seedance-2.5` is the long-form, multimodal tier**, and the only model that does either of the two things below. Reach for it when the beat cannot be five seconds, or when the source is footage rather than a still.
 
@@ -3512,7 +3512,7 @@ baker studio animate "she keeps walking, camera stays with her, then she stops a
 | `--from` | `frame` (default — animate the first image) \| `references` (use the assets as visual guidance) \| `text` (no image; render the prompt). The first two are mutually exclusive at every provider |
 | `--from-clip` | Studio takes to build on, `batchId:slot` comma-separated (Seedance 2.5, needs `--from references`) |
 | `--from-video` | Library video ids to build on — same 10-clip budget as `--from-clip` |
-| `--model` | Video model id (default `google/gemini-omni-flash`) |
+| `--model` | Video model id (default `google/gemini-omni-1.1-flash`) |
 | `--aspect-ratio` | Default `9:16` |
 | `--duration` | Seconds; per-model set (default 5, snapped onto what the model renders) |
 | `--quality` | Resolution, e.g. `720p` (default) \| `1080p` |
@@ -4316,7 +4316,7 @@ Fail-fast: an invalid model selection (wrong count, duplicate or out-of-range pi
 
 ##### `video_generate`
 
-Generate video. Curated model roster (per-model params below). Async with polling, except `google/gemini-omni-flash`, which returns its clip inline.
+Generate video. Curated model roster (per-model params below). Async with polling, except `google/gemini-omni-1.1-flash`, which returns its clip inline.
 
 **Inputs**
 
@@ -4339,20 +4339,23 @@ Accepted ref-image MIMEs vary by model — see per-model sections below.
 
 ---
 
-###### Model: `google/gemini-omni-flash` (default)
+###### Model: `google/gemini-omni-1.1-flash` (default)
 
-The default `video_generate` model (`DEFAULT_VIDEO_GENERATE_MODEL`). Natively multimodal — one call returns picture **and** synchronized audio. **Not on OpenRouter** (its `GET /api/v1/videos/models` catalogue of 17 video models does not include it): routed direct to Google's **Interactions API**, `POST https://generativelanguage.googleapis.com/v1beta/interactions`, wire model `gemini-omni-flash-preview`. Its own `generateContent` endpoint refuses the model ("This model only supports Interactions API"), and it does not use Veo's `predictLongRunning` shape. The call is synchronous (~30–40s) and returns the MP4 inline as base64 — no polling.
+The default `video_generate` model (`DEFAULT_VIDEO_GENERATE_MODEL`). Natively multimodal — one call returns picture **and** synchronized audio. **Not on OpenRouter** (its `GET /api/v1/videos/models` catalogue of 17 video models does not include it): routed direct to Google's **Interactions API**, `POST https://generativelanguage.googleapis.com/v1beta/interactions`, wire model `gemini-omni-1.1-flash`. Its own `generateContent` endpoint refuses the model ("This model only supports Interactions API"), and it does not use Veo's `predictLongRunning` shape. The call is synchronous (~30–40s) and returns the MP4 inline as base64 — no polling.
 
 Ref-image MIMEs: `image/png`, `image/jpeg`, `image/webp`. Takes **one** conditioning image — there is no `last_frame`, and wiring one raises `VIDEO_MODEL_IGNORES_INPUT`. Tasks: `text_to_video`, `image_to_video`, `reference_to_video`, `edit`, `extend` — the only curated model besides Seedance that accepts `references`.
 
 | Name | Type | Required | Notes |
 |---|---|---|---|
-| `model` | literal | yes | `"google/gemini-omni-flash"` |
+| `model` | literal | yes | `"google/gemini-omni-1.1-flash"` |
 | `prompt` | string | yes | non-empty |
 | `aspect_ratio` | enum | no | `16:9 \| 9:16` (`1:1` is rejected) |
+| `resolution` | enum | no | `360p \| 720p \| 1080p \| 4k`. Omitted ⇒ **720p**. Note the lowercase `4k` — `4K` is Veo's spelling and is rejected here. |
 | `duration` | enum | no | `3, 4, 5, 6, 7, 8, 9, 10` seconds. Omitted ⇒ **10s** — the model's own default *and* its ceiling. |
 
-> Those are the **only** accepted params. `resolution`, `seed`, `negative_prompt`, `generate_audio`, `person_generation`, `enhance_prompt` and `conditioning_scale` each fail pre-dispatch with a 400 `Unknown parameter`. Output is fixed: **720p** (1280×720 or 720×1280), 24fps, h264 + native stereo AAC — audio is always on.
+> Those are the **only** accepted params. `seed`, `negative_prompt`, `generate_audio`, `person_generation`, `enhance_prompt` and `conditioning_scale` each fail pre-dispatch with a 400 `Unknown parameter`. Output is 24fps, h264 + native stereo AAC — audio is always on and has no toggle. Measured frame sizes: 640×360, 1280×720, 1920×1080, 3840×2160 (the last two are upscales of the same render).
+
+> **`resolution` is new in 1.1** — the preview model had no such knob and 400'd the field, which is why older notes say Omni is 720p-only. **Video input is a different story: it does not work on our key.** 1.1 advertises video references, `edit` and `extend`, and every request carrying a video input part comes back 400 `content_blocked`, whether the clip is inlined as base64 or uploaded to the Files API and passed by `uri` — including a clip Omni itself had just rendered. An ordinary `https://` URL is refused separately with "HTTP video input is not supported by the model". Image references on the same key and request shape render fine. Until that clears, a beat that must build on existing footage belongs on `bytedance/seedance-2.5`.
 
 > `duration` is the exception. Omni has **no duration field** — every spelling at every nesting level (`duration`, `duration_seconds`, `seconds`, `length`, `video_length`, `num_frames`, at top level, inside `response_format`, inside `generation_config.video_config`) comes back `Unknown parameter`. But it **honours a length asked for in the prompt**, so the backend adapter (`convex/lib/geminiOmni.ts`) translates the canonical `duration` param into a prompt directive. Measured: "exactly 3 seconds" → 3.008s, "exactly 6 seconds" → 6.016s, no length named → 10.005s.
 
@@ -4361,13 +4364,30 @@ Ref-image MIMEs: `image/png`, `image/jpeg`, `image/webp`. Takes **one** conditio
 ```json
 { "id": "clip", "type": "video_generate",
   "inputs": { "first_frame": "$ref:start.images#0" },
-  "params": { "model": "google/gemini-omni-flash", "prompt": "Same person, talking head, …",
+  "params": { "model": "google/gemini-omni-1.1-flash", "prompt": "Same person, talking head, …",
               "aspect_ratio": "9:16", "duration": 6 } }
 ```
 
-> **Billed per token, which resolves to a clean per-second rate** — $17.50 per 1M video output tokens ($1.50 per 1M input, noise), and 720p video is a flat **5,792 tokens per second at every length** (3.008s → 17,376 tokens; 6.016s → 34,752; 10.005s → 57,920). So **$0.10136 per second**: a 10s clip ~$1.01 (102 credits), a 4s clip ~$0.41 (41 credits), a 3s clip ~$0.30 (31 credits). Cutting a clip short is a real saving. A 5s 720p Seedance clip is $2.50.
+> **Billed per token, which resolves to a clean per-second rate at each tier** — $17.50 per 1M video output tokens ($1.50 per 1M input, noise). Video output is flat per second of footage and linear in both length and tier, measured: 720p returned 17,376 tokens for 3.008s and 34,752 for 6.016s, and at a fixed 3s the four tiers returned 5,793 / 17,376 / 26,064 / 52,128 — exactly ⅓, 1, 1.5 and 3 times the 720p rate.
+>
+> | Tier | Tokens/s | USD/s | 3s clip | 10s clip |
+> |---|---|---|---|---|
+> | `360p` | 1,931 | $0.0338 | ~$0.10 (11 cr) | ~$0.34 (35 cr) |
+> | `720p` | 5,792 | $0.1014 | ~$0.30 (31 cr) | ~$1.01 (102 cr) |
+> | `1080p` | 8,688 | $0.1520 | ~$0.46 (46 cr) | ~$1.52 (153 cr) |
+> | `4k` | 17,376 | $0.3041 | ~$0.91 (92 cr) | ~$3.04 (305 cr) |
+>
+> Cutting a clip short is a real saving, and so is drafting at `360p`: five 360p takes cost less than two at 720p. (The **preview** model billed its whole 10s window whatever length it returned — a flat ~$1.03 — so anything quoting Omni at a flat rate predates 1.1.) A 5s 720p Seedance clip is $2.50.
+
+> **Move off the default** when a beat needs a length outside 3–10s, a `1:1` frame, existing footage as an input, or a photoreal human face on camera (that still routes to Veo, unchanged). 1080p no longer forces a move.
+
+###### Model: `google/gemini-omni-flash` (legacy preview)
+
+The Omni **preview** (wire model `gemini-omni-flash-preview`), superseded by `google/gemini-omni-1.1-flash`. Kept in the roster so canvases authored before 1.1 still validate and re-run verbatim — **don't pick it for new work.** It is 1.1 minus the quality dial: same per-second rate, `prompt` / `aspect_ratio` / `duration` and nothing else.
+
+> **It does not reject a `resolution` — it ignores one.** Asked for `360p` it returns HTTP 200, a **1280×720** clip, and a bill of **17,376 video tokens** (the 720p rate, 3× the 5,793 that tier costs on 1.1). Accepted, ignored, charged for. That is why the registry declares no `resolution` here and the studio refuses one outright rather than passing it through: this is the one model where asking for a cheaper tier silently costs full price.
 
-> **Move off the default** when a beat needs 1080p, a length outside 3–10s, a `1:1` frame, or a photoreal human face on camera (that still routes to Veo, unchanged).
+> Historical note: this model used to bill its **whole 10s window** whatever length it rendered (a 3s and a 4s clip both ~$1.034). Re-measured, that is gone — 3s bills 17,376 tokens and 6s 34,752, the same per-second line as 1.1.
 
 ###### Model: `bytedance/seedance-2.0`
 
@@ -5050,7 +5070,7 @@ baker canvas run ./reference-ad.video.canvas.json
 | `--deconstruct-model <id>` | `~google/gemini-pro-latest` | Override the `video_deconstruct` model. |
 | `--select-model <id>` | `~google/gemini-flash-latest` | Override the element-selection `text_generate` model. |
 | `--image-model <id>` | `openai/gpt-image-2` | Override the per-frame `image_generate` model (defaults to the strongest, matching `scaffold-static-ad`). |
-| `--video-model <id>` | scored router | Override the `video_generate` model **and skip the router**. Curated roster: `google/gemini-omni-flash` (**the default video model** — native audio + reference images, ~$0.101 per second of clip, 720p/`16:9`\|`9:16`, `duration` 3–10s and 10s when unset), `bytedance/seedance-2.0` (workhorse), `google/veo-3.1` (cine ceiling + real-face), `google/veo-3.1-fast` (cheap Veo), `kwaivgi/kling-v3.0-pro` (motion-transfer). |
+| `--video-model <id>` | scored router | Override the `video_generate` model **and skip the router**. Curated roster: `google/gemini-omni-1.1-flash` (**the default video model** — native audio + reference images, `360p`\|`720p`\|`1080p`\|`4k` at ~$0.034\|$0.101\|$0.152\|$0.304 per second, `16:9`\|`9:16`, `duration` 3–10s and 10s when unset), `bytedance/seedance-2.0` (workhorse), `google/veo-3.1` (cine ceiling + real-face), `google/veo-3.1-fast` (cheap Veo), `kwaivgi/kling-v3.0-pro` (motion-transfer). |
 | `--real-face` | auto-detected | Force the Veo route for a photoreal human presenter (dodges the ByteDance real-person filter). **Usually unnecessary** — the scaffold auto-detects a photoreal on-camera person/animal cast from the deconstruct and routes it to Veo on its own; set this only to force it when the cast isn't detected. Covers AI-generated photoreal faces too, not just real ones. |
 | `--motion-transfer` | off | Motion driven from a reference / hyper-dynamic → the router picks Kling. |
 | `--identity` | off | Same character/product across clips → the router picks the Seedance workhorse. |

package/dist/{chunk-EKLAHWSF.js → chunk-SJPRTSGY.js}

@@ -1191,6 +1191,7 @@ var SEEDANCE_25_MAX_AUDIO_REFERENCES = 10;
 var KLING_DURATIONS = [3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15];
 var GEMINI_OMNI_DURATIONS = [3, 4, 5, 6, 7, 8, 9, 10];
 var GEMINI_OMNI_ASPECT_RATIOS = ["16:9", "9:16"];
+var GEMINI_OMNI_RESOLUTIONS = ["360p", "720p", "1080p", "4k"];
 var GEMINI_OMNI_IMAGE_MIMES = ["image/png", "image/jpeg", "image/webp"];
 var GROK_IMAGINE_DURATIONS = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10];
 var WAN_27_DURATIONS = [2, 3, 4, 5, 6, 7, 8, 9, 10];
@@ -1224,6 +1225,7 @@ var DEFAULT_IMAGE_GENERATE_MODEL = "google/gemini-3.1-flash-image-preview";
 var VIDEO_GENERATE_MODELS = [
   "bytedance/seedance-2.0",
   "bytedance/seedance-2.5",
+  "google/gemini-omni-1.1-flash",
   "google/gemini-omni-flash",
   "google/veo-3.1",
   "google/veo-3.1-fast",
@@ -1233,10 +1235,11 @@ var VIDEO_GENERATE_MODELS = [
   "alibaba/wan-2.7",
   "alibaba/happyhorse-1.1"
 ];
-var DEFAULT_VIDEO_GENERATE_MODEL = "google/gemini-omni-flash";
+var DEFAULT_VIDEO_GENERATE_MODEL = "google/gemini-omni-1.1-flash";
 var REFERENCE_TO_VIDEO_MODELS = [
   "bytedance/seedance-2.0",
   "bytedance/seedance-2.5",
+  "google/gemini-omni-1.1-flash",
   "google/gemini-omni-flash",
   "x-ai/grok-imagine-video",
   "alibaba/wan-2.7",
@@ -1248,6 +1251,10 @@ function supportsReferenceToVideo(model) {
 function isGeminiOmniModel(model) {
   return model.startsWith("google/gemini-omni");
 }
+var GEMINI_OMNI_PREVIEW_MODEL = "google/gemini-omni-flash";
+function isGeminiOmniPreviewModel(model) {
+  return model === GEMINI_OMNI_PREVIEW_MODEL;
+}
 function isSeedance25Model(model) {
   return model === "bytedance/seedance-2.5";
 }
@@ -1603,32 +1610,87 @@ var MODEL_REGISTRY = {
         watermark: { kind: "boolean" }
       }
     },
-    "google/gemini-omni-flash": {
-      // Google Gemini Omni Flash — the DEFAULT video model. Natively multimodal:
-      // it reasons over text + images and renders video WITH synchronized audio in
-      // one call, and it is the only curated model that takes reference images
-      // through its own first-party API (task `reference_to_video`).
+    "google/gemini-omni-1.1-flash": {
+      // Google Gemini Omni 1.1 Flash — the DEFAULT video model. Natively
+      // multimodal: it reasons over text + images and renders video WITH
+      // synchronized audio in one call, and it is the only curated model that
+      // takes reference images through its own first-party API (task
+      // `reference_to_video`).
+      //
+      // It is a SEPARATE entry from `google/gemini-omni-flash` (the preview),
+      // not a rolling upgrade of it, because the two do not render the same
+      // things: 1.1 takes a `resolution` and the preview does not, and 1.1 bills
+      // the seconds it renders where the preview bills its whole 10s window. One
+      // shared key would have to promise the union of those, which means a canvas
+      // could ask the preview for 360p and be quoted a third of what it charges.
       //
       // NOT on OpenRouter. `GET /api/v1/videos/models` lists 17 video models and
-      // Omni is not among them, so this model is the sanctioned direct-provider
+      // no Gemini Omni at either version, so this is the sanctioned direct-provider
       // exception (see `convex/lib/geminiOmni.ts`): Google's Interactions API,
       // `POST /v1beta/interactions`. Its own `generateContent` endpoint refuses it
       // ("This model only supports Interactions API"), so the Veo `predictLongRunning`
       // shape does not apply either.
       //
-      // The param set below is DELIBERATELY tiny and was read off the live API, not
-      // off the docs — every knob the other video models take is rejected with a
-      // pre-dispatch 400 (`Unknown parameter '<name>'`): no `resolution`, no `seed`,
-      // no `negative_prompt`, no `generate_audio`, no `person_generation` /
-      // `enhance_prompt` / `conditioning_scale`. Resolution is fixed at 720p/24fps
-      // and audio is always on. Widening this list does not widen the model; it just
-      // moves a free validation error into a billed one.
+      // The param set below is deliberately short and was read off the live API,
+      // not off the docs — the knobs absent here are rejected with a pre-dispatch
+      // 400 (`Unknown parameter '<name>'`): no `seed`, no `negative_prompt`, no
+      // `generate_audio`, no `person_generation` / `enhance_prompt` /
+      // `conditioning_scale`. Audio is always on and cannot be turned off.
+      // Widening this list does not widen the model; it just moves a free
+      // validation error into a billed one.
+      //
+      // `duration` is the odd one out: Omni has no duration FIELD at any nesting
+      // level, but it honours a length named in the prompt (3-10s), so the adapter
+      // turns this param into a prompt directive. It is a real, billable knob.
+      //
+      // NO VIDEO INPUT SLOT, and that is a gate on our project rather than a gap
+      // in the model or a mistake in how we ask. 1.1 advertises video references,
+      // `edit` and `extend`, and every request carrying a video comes back 400
+      // `content_blocked`.
+      //
+      // The controlled version, run through the shape Google's own docs use
+      // (`{"type":"document","uri":<Files API uri>}`, no generation_config),
+      // changing only the uploaded file: an IMAGE returns 200 and renders a clip;
+      // a VIDEO returns `content_blocked`. Same route, same structure. So the
+      // request shape is right and video input specifically is refused. Also
+      // measured across inline base64, all three tasks, video AND text output,
+      // synthetic clips, Omni's own output, real camera footage, and both our
+      // keys. Full probe log: `.context/omni-1.1-measurements.md`.
       //
-      // `duration` is the exception, and the reason it looks odd: Omni has no
-      // duration FIELD, but it honours a length named in the prompt (3-10s), so the
-      // adapter turns this param into a prompt directive. It is a real, billable
-      // knob — cost scales exactly with the seconds rendered.
-      label: "Google Gemini Omni Flash",
+      // Adding `{ kind: "video", mimes: … }` below is the ONE line that turns the
+      // whole feature on — the "Build on" control, the reference picker taking an
+      // MP4, the drop target and the source-clip cost premium are all driven off
+      // this slot. Add it the day the 400 stops, not before: until then it offers
+      // a control whose every use is refused.
+      label: "Google Gemini Omni 1.1 Flash",
+      inputs: [],
+      optional_inputs: [{ kind: "image", mimes: GEMINI_OMNI_IMAGE_MIMES }],
+      required: ["prompt"],
+      params: {
+        prompt: { kind: "string" },
+        aspect_ratio: { kind: "string", enum: GEMINI_OMNI_ASPECT_RATIOS },
+        resolution: { kind: "string", enum: GEMINI_OMNI_RESOLUTIONS },
+        duration: { kind: "number", enum: GEMINI_OMNI_DURATIONS }
+      }
+    },
+    "google/gemini-omni-flash": {
+      // LEGACY — the Omni PREVIEW (`gemini-omni-flash-preview`), superseded by
+      // `google/gemini-omni-1.1-flash` above. Kept so canvases authored before 1.1
+      // still validate and re-run unchanged (`baker canvas rerun` restores a
+      // recorded canvas verbatim, so dropping this model would break every stored
+      // creative that used it). Don't reach for it in new work.
+      //
+      // Its param surface is the ORIGINAL one on purpose, and the omission that
+      // matters is `resolution`. This model renders 720p / 24fps and nothing else.
+      //
+      // It does NOT reject the field, which is the trap and the reason this is
+      // enforced here rather than left to the provider: asked for `360p` it
+      // returns HTTP 200, a 1280x720 clip, and a bill of 17,376 video tokens —
+      // the 720p rate, three times the 5,793 that tier costs on 1.1. Accepted,
+      // ignored, and charged for. Copying 1.1's enum across to make the two
+      // entries match would hand a canvas a quality dial that does nothing and a
+      // quote a third of the real charge.
+      label: "Google Gemini Omni Flash (legacy preview)",
       inputs: [],
       optional_inputs: [{ kind: "image", mimes: GEMINI_OMNI_IMAGE_MIMES }],
       required: ["prompt"],
@@ -2218,9 +2280,20 @@ var USD_PER_SECOND = {
 };
 var GEMINI_OMNI_USD_PER_1M_VIDEO_TOKENS = 17.5;
 var GEMINI_OMNI_VIDEO_TOKENS_PER_SECOND = 5792;
+var GEMINI_OMNI_RESOLUTION_MULTIPLIER = {
+  "360p": 1 / 3,
+  "720p": 1,
+  "1080p": 1.5,
+  "4k": 3
+};
+var GEMINI_OMNI_DEFAULT_RESOLUTION = "720p";
+var GEMINI_OMNI_OVERHEAD_TOKENS = 600;
 var GEMINI_OMNI_WINDOW_SECONDS = 10;
-var GEMINI_OMNI_WINDOW_MARGIN = 1.05;
 var GEMINI_OMNI_USD_PER_SECOND = GEMINI_OMNI_VIDEO_TOKENS_PER_SECOND * GEMINI_OMNI_USD_PER_1M_VIDEO_TOKENS / 1e6;
+function geminiOmniUsdPerSecond(resolution) {
+  const multiplier = GEMINI_OMNI_RESOLUTION_MULTIPLIER[resolution ?? GEMINI_OMNI_DEFAULT_RESOLUTION] ?? 1;
+  return GEMINI_OMNI_USD_PER_SECOND * multiplier;
+}
 var VIDEO_SOURCE_COST_MULTIPLIER = 2.5;
 var FAL_SEEDANCE_USD_PER_1K_TOKENS = 0.0214;
 var SEEDANCE_SOURCE_CLIP_DISCOUNT = 0.6;
@@ -2234,13 +2307,16 @@ function falSeedanceUsdPerSecond(resolution, buildsOnSourceClip = false) {
   const rate = tokensPerSecond * FAL_SEEDANCE_USD_PER_1K_TOKENS / 1e3;
   return buildsOnSourceClip ? rate * SEEDANCE_SOURCE_CLIP_DISCOUNT : rate;
 }
+function geminiOmniCostUsd(model, duration, resolution) {
+  const tier = isGeminiOmniPreviewModel(model) ? GEMINI_OMNI_DEFAULT_RESOLUTION : resolution;
+  const footage = geminiOmniUsdPerSecond(tier) * (duration ?? GEMINI_OMNI_WINDOW_SECONDS);
+  return footage + GEMINI_OMNI_OVERHEAD_TOKENS * GEMINI_OMNI_USD_PER_1M_VIDEO_TOKENS / 1e6;
+}
 function estimateVideoCostUsd(input) {
   const { model, duration, resolution, generateAudio, buildsOnSourceClip } = input;
   const seconds = billedSeconds(model, duration ?? DEFAULT_VIDEO_DURATION_S);
   const sourceClipPremium = buildsOnSourceClip ? VIDEO_SOURCE_COST_MULTIPLIER : 1;
-  if (isGeminiOmniModel(model)) {
-    return GEMINI_OMNI_WINDOW_SECONDS * GEMINI_OMNI_USD_PER_SECOND * GEMINI_OMNI_WINDOW_MARGIN;
-  }
+  if (isGeminiOmniModel(model)) return geminiOmniCostUsd(model, duration, resolution);
   if (isSeedance25Model(model)) {
     const viaOpenRouter = seedance25UsdPerSecond(resolution) * seconds * sourceClipPremium;
     if (!input.mayRouteToFal) return viaOpenRouter;
@@ -3137,9 +3213,9 @@ function clipProfileFor(modelId) {
   if (/^kwaivgi\/kling|^kling\//.test(modelId)) return KLING_PROFILE;
   return void 0;
 }
-function clipParamRecipe(profile, intent) {
+function clipParamRecipe(model, profile, intent) {
   const out = {};
-  if (intent === "hero" && profile.id !== "omni") {
+  if (intent === "hero" && videoModelRenders(model, "1080p")) {
     out.resolution = "1080p";
   }
   if (intent === "hook" && profile.id === "kling") {
@@ -3147,6 +3223,10 @@ function clipParamRecipe(profile, intent) {
   }
   return out;
 }
+function videoModelRenders(model, resolution) {
+  const schema = MODEL_REGISTRY.video_generate[model]?.params.resolution;
+  return schema?.kind === "string" && (schema.enum ?? []).includes(resolution);
+}
 function nativeDialogueOf(prompt) {
   if (typeof prompt !== "string") return void 0;
   const m = prompt.match(/Dialogue: "(.*)"/);
@@ -8537,8 +8617,8 @@ var videoGenerateNode = delegated({
   id: "video_generate",
   version: "2.0.0",
   category: "video",
-  summary: "Generate video for ad creatives. Curated roster: `google/gemini-omni-flash` (DEFAULT \u2014 native audio + reference images in one call, 720p, 3-10s), `bytedance/seedance-2.0` (identity/product workhorse, NO photoreal human faces), `google/veo-3.1` (photoreal cine ceiling + the human-face model), `google/veo-3.1-fast` (cheap Veo iteration), `kwaivgi/kling-3.0` (motion-transfer/dynamic). Async with polling.",
-  when_to_use: "`google/gemini-omni-flash` is the DEFAULT: it renders picture and synchronized audio in one call, and is the only model here that takes reference images alongside a prompt. SET `duration` (3-10s) to the length the beat actually needs \u2014 it bills ~$0.101 PER SECOND, so leaving it out renders the 10s default and charges ~$1.01 for a beat that may need 3s (~$0.30). Its resolution is fixed at 720p and it accepts only `16:9` or `9:16`; it takes NO `resolution`, `seed`, `negative_prompt` or `generate_audio` param \u2014 sending one fails validation. Keep the default for ordinary beats; move off it when a beat needs 1080p, a length outside 3-10s, or a 1:1 frame. Use `bytedance/seedance-2.0` for identity/product output WITHOUT a photoreal human face on camera. Any clip whose frame shows a photoreal human face \u2014 a UGC creator, presenter, or testimonial speaker, REAL OR AI-GENERATED \u2014 must use `google/veo-3.1`: ByteDance's filter rejects photoreal human faces on Seedance (content_policy_blocked/E005), and a synthetic AI face does NOT clear it. Use `google/veo-3.1-fast` while iterating to keep cost low; `kwaivgi/kling-3.0` for motion-transfer/hyper-dynamic beats. `baker canvas scaffold-video` auto-routes a photoreal cast to Veo; a Seedance clip blocked at runtime also auto-recovers on Veo. Each model gates its own durations/resolutions/aspect ratios in the registry \u2014 see the README per-model section.",
+  summary: "Generate video for ad creatives. Curated roster: `google/gemini-omni-1.1-flash` (DEFAULT \u2014 native audio + reference images in one call, 360p-4k, 3-10s), `bytedance/seedance-2.0` (identity/product workhorse, NO photoreal human faces), `google/veo-3.1` (photoreal cine ceiling + the human-face model), `google/veo-3.1-fast` (cheap Veo iteration), `kwaivgi/kling-3.0` (motion-transfer/dynamic). Async with polling.",
+  when_to_use: "`google/gemini-omni-1.1-flash` is the DEFAULT: it renders picture and synchronized audio in one call, and is the only model here that takes reference images alongside a prompt. SET `duration` (3-10s) to the length the beat actually needs \u2014 it bills PER SECOND, so leaving it out renders the 10s default and charges for ten seconds of a beat that may need three. SET `resolution` too: `360p` bills a THIRD of `720p` ($0.034 vs $0.101 per second) and is the right tier while you are still choosing between takes \u2014 render the drafts at 360p, then re-render the keeper at `720p` (default), `1080p` ($0.152/s) or `4k` ($0.304/s). It accepts only `16:9` or `9:16`, and takes NO `seed`, `negative_prompt` or `generate_audio` param \u2014 sending one fails validation, and it always renders audio. Keep the default for ordinary beats; move off it only when a beat needs a length outside 3-10s or a 1:1 frame. Use `bytedance/seedance-2.0` for identity/product output WITHOUT a photoreal human face on camera. Any clip whose frame shows a photoreal human face \u2014 a UGC creator, presenter, or testimonial speaker, REAL OR AI-GENERATED \u2014 must use `google/veo-3.1`: ByteDance's filter rejects photoreal human faces on Seedance (content_policy_blocked/E005), and a synthetic AI face does NOT clear it. Use `google/veo-3.1-fast` while iterating to keep cost low; `kwaivgi/kling-3.0` for motion-transfer/hyper-dynamic beats. `baker canvas scaffold-video` auto-routes a photoreal cast to Veo; a Seedance clip blocked at runtime also auto-recovers on Veo. Each model gates its own durations/resolutions/aspect ratios in the registry \u2014 see the README per-model section.",
   inputs: z29.object({
     first_frame: ImageRef.optional(),
     last_frame: ImageRef.optional(),
@@ -9074,4 +9154,4 @@ export {
   defaultRegistry,
   createEngineFromEnv
 };
-//# sourceMappingURL=chunk-EKLAHWSF.js.map
+//# sourceMappingURL=chunk-SJPRTSGY.js.map