@nodaro/prompts 1.1.1 → 1.2.0

package/dist/index.cjs

@@ -26058,8 +26058,45 @@ precise subject \u2192 action details \u2192 scene/environment \u2192 lighting &
 - More than 4 referenced people gets unstable: group people into composite images of \u22644 first (image generation), then reference those composites.
 - Repeated extension degrades quality: prefer high-definition reference assets and avoid stacking many continuations.`
 };
+var KLING_AUDIO_DOCTRINE = {
+  providers: ["kling", "kling-3.0", "kling-3-omni"],
+  heading: "Kling 2.6 / 3.0 / 3 Omni (kling, kling-3.0, kling-3-omni)",
+  tips: [
+    'Kling speaks scripted dialogue natively with lip sync \u2014 quote the line and enable sound: [Anna: warm calm voice]: "We made it." On kling/kling-3.0 audio raises the credit cost; kling-3-omni includes it.',
+    "Structure prompts as Scene \u2192 character/element \u2192 Motion \u2192 Audio \u2192 style. Put ALL sound in one 'Audio:' block: dialogue in quotes, then SFX and ambience described plainly ('rain tapping on glass, no music').",
+    "Give each speaker a stable label + voice description and reuse it exactly \u2014 [Detective: low raspy voice, tired] \u2014 pronouns or renamed speakers break voice binding in multi-character scenes.",
+    "Tone words in the bracket steer delivery (whispering, crying, fast urgent voice); pace with 'Immediately' / 'after a pause'. 2.6 voices are English/Chinese only; 3.0 adds dialects and code-switching.",
+    "kling-3.0 wired references become @element_name mentions (handled automatically by the editor); multi-shot mode forces sound ON. Kling 2.6 prompts cap at 1000 chars \u2014 keep the Audio block tight.",
+    "Say what should NOT sound: 'no background music, no other sounds' \u2014 otherwise Kling invents a music bed under dialogue."
+  ],
+  doctrine: `Prompt structure: Scene (setting, light) \u2192 Character/Element (who, appearance) \u2192 Motion (action, camera) \u2192 Audio (dialogue / SFX / ambience / music) \u2192 Others (style, emotion).
+
+**Dialogue (native speech + lip sync \u2014 verified on the KIE path 2026-07-16)**
+- Quote the spoken line and enable the sound toggle; the model bakes the voice AND matching lip movement: the woman says "The quick brown fox jumps over the lazy dog."
+- Prefer labeled dialogue with a voice description: [Character label: voice/tone description]: "line". Example: [Exhausted Partner: trembling frustrated voice]: "You never listen to me."
+- Keep character labels unique and reuse them verbatim \u2014 never switch to pronouns mid-prompt; the label is what binds a voice to a speaker across lines. Kling 2.6 additionally supports [Character@VoiceName] platform-voice binding.
+- Tone words inside the bracket steer delivery: whispering, crying voice, controlled serious voice, fast urgent voice. Sequence speech with temporal markers ("Immediately", "after a pause") when two lines must not overlap.
+- Languages: Kling 2.6 outputs English/Chinese voices only (other languages are auto-translated to English). Kling 3.0 supports multiple languages, dialects, accents, and code-switching within one scene \u2014 mark the language explicitly ("says in Japanese \u2026").
+
+**SFX / ambience / music**
+- Put them in the same Audio block, described plainly: "Rain tapping softly on the window, distant thunder, no music."
+- State exclusions explicitly \u2014 "no background music, no other sounds" \u2014 or the model tends to add a bed under dialogue.
+
+**Toggle + cost**
+- The audio lever is the node's sound toggle (KIE \`sound\` param). On kling (2.6) and kling-3.0 enabling audio raises the credit cost (the \`:audio\` composite); kling-3.0 generates audio by DEFAULT \u2014 pass sound: false for the cheaper silent tier. kling-3-omni (Replicate) includes audio in its flat per-duration rate.
+- Multi-shot kling-3.0 (\`multi_shots\`) forces sound ON \u2014 budget for the audio rate.
+
+**References & elements (kling-3.0 / omni)**
+- Wired references are injected as \`kling_elements\` and MUST be mentioned as @element_name in the prompt \u2014 the editor's {image:N} tokens and the server prefixer handle this automatically; when hand-writing prompts, mention every element or it is silently ignored.
+- kling-3-omni is image-to-video only (start frame required) and accepts up to 7 reference images; element voice references (element_input_audio_urls, 5-30s clips) bind a voice to an element.
+
+**Limits**
+- Kling 2.6 prompts cap at 1000 characters \u2014 front-load scene + dialogue and trim style tails first. kling-3.0 accepts long prompts.
+- Durations: 2.6 = 5/10s; 3.0/omni = 3-15s. A spoken line needs roughly 1s per 2-3 words \u2014 don't script more dialogue than the clip can hold.`
+};
 var PROVIDER_PROMPT_DOCTRINES = [
-  SEEDANCE_2_DOCTRINE
+  SEEDANCE_2_DOCTRINE,
+  KLING_AUDIO_DOCTRINE
 ];
 var DOCTRINE_BY_PROVIDER = new Map(
   PROVIDER_PROMPT_DOCTRINES.flatMap((d) => d.providers.map((p) => [p, d]))
@@ -26159,6 +26196,7 @@ var PROVIDER_CAPABILITIES = {
     "nano-banana": "Fast generation, style flexibility, reference image support",
     "nano-banana-pro": "Higher quality Nano Banana with better detail",
     "nano-banana-2": "Latest Nano Banana with resolution options (1K/2K/4K)",
+    "nano-banana-2-lite": "Fast, low-cost 1K generation for drafts and iteration",
     "gpt-image": "Creative concepts, illustration, variable quality tiers",
     "gpt-image-2": "Latest GPT Image \u2014 sharp text, photorealism, 1K/2K/4K resolution",
     "grok": "General purpose, good text understanding",
@@ -26169,6 +26207,7 @@ var PROVIDER_CAPABILITIES = {
     "qwen": "Versatile, good prompt adherence",
     "seedream": "Artistic, painterly styles, creative interpretation",
     "seedream-5-lite": "Lighter Seedream, faster artistic generation",
+    "seedream-5-pro": "Flagship Seedream \u2014 strongest instruction following, 2K via high quality",
     "z-image": "Experimental, novel generation approaches",
     "wan-2.7": "Wan 2.7 T2I \u2014 1K/2K/4K, up to 9 ref images",
     "wan-2.7-pro": "Wan 2.7 Pro T2I \u2014 higher quality, 1K/2K/4K",
@@ -26191,6 +26230,7 @@ var PROVIDER_CAPABILITIES = {
     "qwen-edit": "Instruction-based editing",
     "seedream-edit": "Artistic style editing",
     "seedream-5-lite-i2i": "Light artistic transformation",
+    "seedream-5-pro-i2i": "Pro-grade instruction-based edits, multi-reference",
     "flux-kontext": "Character-consistent edits with reference awareness",
     "flux-kontext-max": "Premium character-consistent editing",
     "kontext-multi": "Multi-image Kontext via Replicate \u2014 up to 4 refs, no safety filter",
@@ -26213,6 +26253,7 @@ var PROVIDER_CAPABILITIES = {
     "qwen-edit": "Instruction-based editing",
     "seedream-edit": "Artistic style editing",
     "seedream-5-lite-i2i": "Light artistic transformation",
+    "seedream-5-pro-i2i": "Pro-grade instruction-based edits, multi-reference",
     "flux-kontext": "Character-consistent edits with reference awareness",
     "flux-kontext-max": "Premium character-consistent editing",
     "kontext-multi": "Multi-image Kontext via Replicate \u2014 up to 4 refs, no safety filter",