@nodaro/prompts 1.55.0 → 1.56.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nodaro/prompts",
-  "version": "1.55.0",
+  "version": "1.56.0",
   "description": "Nodaro's prompt-engineering layer — person/picker catalogs with prompt hints, identity-lock clauses, entity prompt builders, brand presets, and prompt/reference assembly shared by the Nodaro platform and SDK.",
   "type": "module",
   "license": "FSL-1.1-Apache-2.0",
@@ -70,7 +70,7 @@
     "test": "vitest run"
   },
   "dependencies": {
-    "@nodaro/shared": "^3.63.0"
+    "@nodaro/shared": "^3.64.0"
   },
   "devDependencies": {
     "tsup": "^8.5.0",

package/src/__tests__/assemble-image-input-style-references.test.ts

@@ -171,3 +171,27 @@ describe("styleReferences — fill the free seats, after the caller's own", () =
     expect(() => assembleImageInput({ userPrompt: "", provider: "nano-banana-pro", throwOnEmpty: true })).toThrow(/No prompt/)
   })
 })
+
+describe("styleReferences — a seat another image of the run takes (`reservedImageSeats`)", () => {
+  it("fills one seat fewer when a base image takes one, so the run never goes past the model's limit", () => {
+    const cap = imageReferenceLimit("flux-3-klein")
+    const reserved = assembleImageInput({
+      userPrompt: "a cat",
+      provider: "flux-3-klein",
+      extraReferenceImageUrls: flat(2),
+      styleReferences: style(cap),
+      reservedImageSeats: 1,
+    })
+    expect(reserved.referenceImageUrls).toHaveLength(cap - 1)
+    expect(reserved.styleReferencesSent).toBe(cap - 3)
+  })
+
+  it("never removes the caller's own references, and 0 changes nothing", () => {
+    const cap = imageReferenceLimit("flux-3-klein")
+    const full = assembleImageInput({ userPrompt: "a cat", provider: "flux-3-klein", extraReferenceImageUrls: flat(cap), styleReferences: style(3), reservedImageSeats: 1 })
+    expect(full.referenceImageUrls).toEqual(flat(cap))
+    expect(full.styleReferencesSent).toBe(0)
+    const input = { userPrompt: "a cat", provider: "nano-banana-pro", extraReferenceImageUrls: flat(2), styleReferences: style(3) }
+    expect(assembleImageInput({ ...input, reservedImageSeats: 0 })).toEqual(assembleImageInput(input))
+  })
+})

package/src/__tests__/seedance-extend.test.ts

@@ -59,3 +59,34 @@ describe("Kling 4 extend shared wiring", () => {
     expect(PROVIDER_CAPABILITIES["extend-video"]!["kling-4-extend"]).toMatch(/ANY video.*Kling 4/i)
   })
 })
+
+describe("PixVerse V6 extend shared wiring", () => {
+  it("is a registered extend provider with its own window and qualities", () => {
+    expect(EXTEND_VIDEO_PROVIDERS).toContain("pixverse-v6-extend")
+    const m = MODEL_CATALOG["pixverse-v6-extend"]!
+    expect(m.modes).toEqual(["extend"])
+    expect(m.family).toBe("PixVerse")
+    expect(m.durations).toEqual([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15])
+    expect(m.resolutions).toEqual(["360p", "540p", "720p", "1080p"])
+  })
+
+  it("is advertised to the wizard's extend-video capabilities", () => {
+    expect(PROVIDER_CAPABILITIES["extend-video"]!["pixverse-v6-extend"]).toMatch(/ANY video.*PixVerse V6/i)
+  })
+})
+
+describe("Grok extend shared wiring", () => {
+  it("is a registered extend provider adding 6 or 10 s, with no resolution lever (the source's is used)", () => {
+    expect(EXTEND_VIDEO_PROVIDERS).toContain("grok-extend")
+    const m = MODEL_CATALOG["grok-extend"]!
+    expect(m.modes).toEqual(["extend"])
+    expect(m.family).toBe("xAI")
+    expect(m.durations).toEqual([6, 10])
+    expect(m.resolutions).toBeUndefined()
+  })
+
+  it("is advertised to the wizard's extend-video capabilities, as extending Grok Imagine videos only", () => {
+    expect(PROVIDER_CAPABILITIES["extend-video"]!["grok-extend"]).toMatch(/Grok Imagine video.*Grok/i)
+    expect(PROVIDER_CAPABILITIES["extend-video"]!["grok-extend"]).not.toMatch(/ANY video/i)
+  })
+})

package/src/assemble-image-input.ts

@@ -131,6 +131,14 @@ export interface AssembleImageInput {
    * many went. Each carries a short style line (`styleReferenceDescription`).
    */
   styleReferences?: readonly StyleReferenceInput[]
+  /**
+   * Seats of the model's reference limit that another image the run sends
+   * takes, outside this assembly: a base image the job sends first, on a model
+   * that counts it against the same limit (FLUX 3 Klein). Style references fill
+   * only the seats left after them, so the platform never adds an image the
+   * model cannot take. 0 (the default) changes nothing.
+   */
+  reservedImageSeats?: number
   /**
    * Flat subject ids (Person / Styling / props) → folded into the prompt AHEAD
    * of the direction clauses: the subject is the noun phrase the cinematography
@@ -367,7 +375,9 @@ function fillStyleReferences(
 ): AssembleImageResult {
   const base = assembleOnce(input)
   const own = base.referenceImageUrls ?? []
-  const seats = imageReferenceLimit(input.provider) - own.length
+  const asked = Number(input.reservedImageSeats ?? 0)
+  const reserved = Number.isFinite(asked) ? Math.max(0, Math.trunc(asked)) : 0
+  const seats = imageReferenceLimit(input.provider) - own.length - reserved
   const seen = new Set(own)
   const candidates: StyleReferenceInput[] = []
   for (const s of style) {

package/src/prompt-wizard-categories.ts

@@ -188,6 +188,7 @@ export const PROVIDER_CAPABILITIES: Record<string, Record<string, string>> = {
     "wan-2.7":     "Wan 2.7 T2I — 1K/2K/4K, up to 9 ref images",
     "wan-2.7-pro": "Wan 2.7 Pro T2I — higher quality, 1K/2K/4K",
     "flux-2-klein": "Open Flux 2 9B via Replicate — fast",
+    "flux-3-klein": "FLUX 3 Klein — the light FLUX 3 model: fast and low cost, 1K/2K/4K; edits from up to 10 references at the same price",
     "flux-2-pro": "BFL Flux 2 Pro via Replicate — flagship quality",
     "flux-2-max": "BFL Flux 2 Max via Replicate — even larger than Pro, up to 8 refs (variable pricing)",
   },
@@ -214,6 +215,7 @@ export const PROVIDER_CAPABILITIES: Record<string, Record<string, string>> = {
     "seedream-5-flash-i2i": "Fast, low-cost Seedream edits — up to 10 input images",
     "minimax-h3-image-i2i": "MiniMax H3 edits — up to 16 reference images in one edit, readable text, up to 4K",
     "seedream-5-5-i2i": "Seedream 5.5 edits — round after round, the rest of the picture untouched; up to 10 input images",
+    "flux-3-klein-i2i": "FLUX 3 Klein edits — fast, low-cost changes from up to 10 reference images, which cost nothing extra; 1K/2K/4K",
     "flux-kontext": "Character-consistent edits with reference awareness",
     "flux-kontext-max": "Premium character-consistent editing",
     "kontext-multi": "Multi-image Kontext via Replicate — up to 4 refs",
@@ -244,6 +246,7 @@ export const PROVIDER_CAPABILITIES: Record<string, Record<string, string>> = {
     "seedream-5-flash-i2i": "Fast, low-cost Seedream edits — up to 10 input images",
     "minimax-h3-image-i2i": "MiniMax H3 edits — up to 16 reference images in one edit, readable text, up to 4K",
     "seedream-5-5-i2i": "Seedream 5.5 edits — round after round, the rest of the picture untouched; up to 10 input images",
+    "flux-3-klein-i2i": "FLUX 3 Klein edits — fast, low-cost changes from up to 10 reference images, which cost nothing extra; 1K/2K/4K",
     "flux-kontext": "Character-consistent edits with reference awareness",
     "flux-kontext-max": "Premium character-consistent editing",
     "kontext-multi": "Multi-image Kontext via Replicate — up to 4 refs",
@@ -348,6 +351,8 @@ export const PROVIDER_CAPABILITIES: Record<string, Record<string, string>> = {
     "seedance-2-5-extend": "Extend ANY video by URL — seamless trim-stitched continuation with audio, 4–30s added (Seedance 2.5)",
     "seedance-2-0-extend": "Extend ANY video by URL — seamless trim-stitched continuation with audio, 4–15s added (Seedance 2.0)",
     "kling-4-extend": "Extend ANY video by URL — seamless trim-stitched continuation with audio, 3–30s added (Kling 4)",
+    "pixverse-v6-extend": "Extend ANY video by URL — native continuation with optional sound, 1–15s added at 360p–1080p, joined onto the original clip (PixVerse V6)",
+    "grok-extend": "Extend a Grok Imagine video made here (by its task id) — native continuation with sound, 6 or 10s added at the video's own resolution (Grok)",
   },
   "speech-to-video": {
     "wan-speech": "Wan 2.2 speech-driven video generation",