@nodaro/prompts 1.7.0 → 1.7.3

package/dist/index.cjs

@@ -10017,6 +10017,9 @@ function renderLens(l) {
   if (l.aperture) bits.push(`f/${l.aperture}`);
   return bits.length > 0 ? `Lens: ${bits.join(", ")}.` : "";
 }
+function identityRefsSentence(firstOrdinal, lastOrdinal) {
+  return firstOrdinal === lastOrdinal ? `${REF_BINDING.ordinal(firstOrdinal)} is an identity reference for this shot's subjects \u2014 match its subject's exact appearance; it is not a frame.` : `${REF_BINDING.ordinal(firstOrdinal)} through ${REF_BINDING.ordinal(lastOrdinal)} are identity references for this shot's subjects \u2014 match each subject's exact appearance; they are not frames.`;
+}
 var REF_BINDING = {
   image: (label, n) => `the ${label} from @image_${n}`,
   video: (label, n) => `the ${label} from @video_${n}`,
@@ -10595,11 +10598,12 @@ function promptBindsFirstFrame(prompt) {
   return /@image_\d+\s+as\s+the\s+(first|opening)\s*(\(first\))?\s*frame/i.test(prompt);
 }
 function resolveSeedance2Inputs(args) {
+  const limits = args.limits ?? shared.SEEDANCE_2_REF_LIMITS;
   const firstFrameUrl = clean(args.firstFrameUrl);
   const lastFrameUrl = clean(args.lastFrameUrl);
   const refImages = cleanList(args.refImageUrls);
-  const refVideos = cleanList(args.refVideoUrls).slice(0, shared.SEEDANCE_2_REF_LIMITS.videos);
-  const refAudios = cleanList(args.refAudioUrls).slice(0, shared.SEEDANCE_2_REF_LIMITS.audio);
+  const refVideos = cleanList(args.refVideoUrls).slice(0, limits.videos);
+  const refAudios = cleanList(args.refAudioUrls).slice(0, limits.audio);
   const hasAnyReference = refImages.length > 0 || refVideos.length > 0 || refAudios.length > 0;
   const canUseStrictMode = !hasAnyReference && (Boolean(firstFrameUrl) || !lastFrameUrl);
   if (canUseStrictMode) {
@@ -10609,7 +10613,7 @@ function resolveSeedance2Inputs(args) {
     return { mode: "first-frame", firstFrameUrl, lastFrameUrl: void 0, referenceImageUrls: [], referenceVideoUrls: [], referenceAudioUrls: [], promptSuffix: "", droppedRefImages: 0 };
   }
   const frameCount = (firstFrameUrl ? 1 : 0) + (lastFrameUrl ? 1 : 0);
-  const userImageSlots = Math.max(0, shared.SEEDANCE_2_REF_LIMITS.images - frameCount);
+  const userImageSlots = Math.max(0, limits.images - frameCount);
   const keptUserImages = refImages.slice(0, userImageSlots);
   const droppedRefImages = refImages.length - keptUserImages.length;
   const referenceImageUrls = [...keptUserImages];
@@ -10643,6 +10647,43 @@ function resolveSeedance2Inputs(args) {
     droppedRefImages
   };
 }
+var GEMINI_OMNI_INPUT_SLOTS = shared.VIDEO_REF_LIMITS_BY_PROVIDER["gemini-omni-video"]?.images ?? 7;
+function resolveGeminiOmniI2vInputs(args) {
+  const refs = (args.refImageUrls ?? []).filter((u) => typeof u === "string" && u.length > 0);
+  const slots = GEMINI_OMNI_INPUT_SLOTS - (args.videoConnected ? 2 : 0);
+  const refSlots = Math.max(0, slots - 1);
+  const kept = refs.slice(0, refSlots);
+  const droppedRefImages = refs.length - kept.length;
+  const imageUrls = [args.firstFrameUrl, ...kept];
+  if (kept.length === 0) return { imageUrls, promptSuffix: "", droppedRefImages };
+  const frameSentence = promptBindsFirstFrame(args.prompt) ? "" : REF_BINDING.frame(1, "opening");
+  const promptSuffix = [frameSentence, identityRefsSentence(2, kept.length + 1)].filter(Boolean).join(" ");
+  return { imageUrls, promptSuffix, droppedRefImages };
+}
+
+// src/veo-i2v-inputs.ts
+var VEO_INGREDIENT_SLOTS = 3;
+function resolveVeoI2vInputs(args) {
+  const refs = (args.refImageUrls ?? []).filter((u) => typeof u === "string" && u.length > 0);
+  if (refs.length === 0) {
+    return {
+      imageUrls: args.endFrameUrl ? [args.firstFrameUrl, args.endFrameUrl] : [args.firstFrameUrl],
+      promptSuffix: "",
+      droppedRefImages: 0,
+      droppedEndFrame: false
+    };
+  }
+  const kept = refs.slice(0, VEO_INGREDIENT_SLOTS - 1);
+  const droppedRefImages = refs.length - kept.length;
+  const frameSentence = promptBindsFirstFrame(args.prompt) ? "" : REF_BINDING.frame(1, "opening");
+  return {
+    imageUrls: [args.firstFrameUrl, ...kept],
+    generationType: "REFERENCE_2_VIDEO",
+    promptSuffix: [frameSentence, identityRefsSentence(2, kept.length + 1)].filter(Boolean).join(" "),
+    droppedRefImages,
+    droppedEndFrame: Boolean(args.endFrameUrl)
+  };
+}
 function toOptions(arr, categoryField) {
   return arr.map((e) => {
     const opt = {
@@ -26224,7 +26265,7 @@ precise subject \u2192 action details \u2192 scene/environment \u2192 lighting &
 **Generation differences (seedance-2-5 vs the 2.0 SKUs)**
 - A single 2.5 shot runs to 30s, where every 2.0 SKU stops at 15s. Plan a complete 4-6 shot beat inside ONE generation instead of splitting it into two clips and stitching \u2014 no seam to hide, and continuity holds because it never leaves the model.
 - 2.5 also takes far more reference material (30 images / 10 videos / 10 audio vs 9/3/3). Treat that as room for COVERAGE \u2014 more distinct characters, locations and props in one shot \u2014 not as licence to pile refs onto one identity. The "ONE headshot + ONE full-body, 4-5 assets total" rule above still produces the best likeness on 2.5.
-- 2.5 renders at 480p/720p only: there is no 1080p or 4K tier, so route a job that needs one to seedance-2 (which has both) or upscale afterwards.
+- 2.5 renders at 480p/720p/1080p (1080p since 2026-08-17): there is no 4K tier, so route a job that needs 4K to seedance-2 (which has it) or upscale afterwards.
 - With a start frame, 2.5 always derives the output aspect from that frame \u2014 an explicit aspect ratio is rejected outright, so compose the frame at the ratio you want.
 
 **References (when reference media is attached)**
@@ -26249,8 +26290,7 @@ 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.
 
-**Auto-path formula (community-sourced enrichment \u2014 apiyi.com Seedance 2.0 prompt guide,
-higgsfield.ai 4K breakdown; captured 2026-08-09)**
+**Auto-path formula (community-sourced enrichment; captured 2026-08-09)**
 - Six steps IN ORDER, 60-100 words total (longer measurably degrades): Subject \u2192 Action \u2192 Environment \u2192 Camera \u2192 Style \u2192 Constraints.
 - ONE primary camera instruction per shot. Compound moves chain with "then": "camera slow tracking then subtle rise" \u2014 never two competing verbs. The 8 reliable camera types: push-in, pull-out, pan, tracking, orbit/arc, aerial, handheld, locked-off.
 - SEPARATE camera movement from subject movement \u2014 the single biggest quality lever: "The dancer spins slowly. Camera holds fixed framing." \u2014 never "spinning camera around a dancing person".
@@ -26643,6 +26683,7 @@ var PROVIDER_CAPABILITIES = {
     "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",
+    "grok-2": "Grok Imagine 2 \u2014 expressive, high-contrast, stylized output",
     "imagen4": "Google's latest, strong photorealism and text rendering",
     "imagen4-fast": "Faster Imagen 4 variant",
     "imagen4-ultra": "Highest quality Imagen 4",
@@ -26719,7 +26760,7 @@ var PROVIDER_CAPABILITIES = {
     "seedance-2": "Seedance 2.0 \u2014 multimodal refs (9 images / 3 videos / 3 audio), native multi-track audio, multi-shot storytelling, 4-15s",
     "seedance-2-fast": "Seedance 2.0 Fast \u2014 same multimodal + audio capabilities, cheaper and quicker",
     "seedance-2-mini": "Seedance 2.0 Mini \u2014 same multimodal + audio capabilities, budget tier, 480p/720p, 4-15s",
-    "seedance-2-5": "Seedance 2.5 \u2014 up to 30s in ONE shot (no stitching), wider multimodal refs (30 images / 10 videos / 10 audio), native audio, 480p/720p",
+    "seedance-2-5": "Seedance 2.5 \u2014 up to 30s in ONE shot (no stitching), wider multimodal refs (30 images / 10 videos / 10 audio), native audio, 480p/720p/1080p",
     "minimax-h3": "MiniMax Hailuo 3 \u2014 premium multimodal refs (9 images / 3 videos / 3 audio), always-on audio, 2K or 768P, 4-15s per-second pricing",
     "wan": "Versatile, good for animations and transformations",
     "wan-turbo": "Faster Wan generation",
@@ -26747,7 +26788,7 @@ var PROVIDER_CAPABILITIES = {
     "seedance-2": "Seedance 2.0 \u2014 start/end frame + multimodal refs, native audio, 4-15s",
     "seedance-2-fast": "Seedance 2.0 Fast \u2014 same capabilities, cheaper and quicker",
     "seedance-2-mini": "Seedance 2.0 Mini \u2014 same capabilities, budget tier, 480p/720p",
-    "seedance-2-5": "Seedance 2.5 \u2014 start/end frame + wide multimodal refs, native audio, up to 30s, 480p/720p",
+    "seedance-2-5": "Seedance 2.5 \u2014 start/end frame + wide multimodal refs, native audio, up to 30s, 480p/720p/1080p",
     "minimax-h3": "MiniMax Hailuo 3 \u2014 first/last frame + multimodal refs, always-on audio, 2K or 768P, 4-15s",
     "hailuo-2.3-pro": "Premium Hailuo animation",
     "hailuo-2.3": "Standard Hailuo animation",
@@ -26860,7 +26901,7 @@ var NODE_PROMPT_CANDIDATE_FIELDS = {
 function computeNodePrompt(nodeType, data, { override, wired, refMap, appendWired }) {
   let typed;
   if (nodeType === "text-to-speech") {
-    typed = data.textSource === "direct" ? [data.directText] : [];
+    typed = data.textSource === "direct" || !present(wired) ? [data.directText] : [];
   } else {
     const fields = NODE_PROMPT_CANDIDATE_FIELDS[nodeType] ?? ["prompt"];
     typed = fields.map((f) => data[f]);
@@ -32075,6 +32116,35 @@ function getPickerWiring(nodeType) {
   return WIRING_MAP.get(nodeType);
 }
 
+// src/surround-fill.ts
+var EDGE = {
+  right: { carried: "left", painted: "right" },
+  left: { carried: "right", painted: "left" },
+  up: { carried: "bottom", painted: "top" },
+  down: { carried: "top", painted: "bottom" }
+};
+var TILT_SUBJECT = {
+  up: {
+    word: "up",
+    subject: "the open sky directly overhead \u2014 sky, clouds, or (for an interior) the canopy or ceiling",
+    where: "overhead"
+  },
+  down: {
+    word: "down",
+    subject: "the ground directly below \u2014 terrain, floor, or water surface",
+    where: "below"
+  }
+};
+function buildSurroundFillPrompt(direction, userPrompt) {
+  const scene = userPrompt && userPrompt.trim() ? `${userPrompt.trim()}. ` : "";
+  const { carried, painted } = EDGE[direction];
+  if (direction === "up" || direction === "down") {
+    const t = TILT_SUBJECT[direction];
+    return `${scene}This is a camera tilted straight ${t.word} from the same scene. The ${carried} strip holds real, finished pixels from the edge of the horizon view; the ${painted} region is flat gray and MUST be painted as ${t.subject}. Render what is genuinely ${t.where} \u2014 do NOT repeat, mirror, or continue the landscape, and do NOT draw a horizon line or distant scenery in the painted region. CRITICAL: keep the ${carried} strip unchanged and match the scene's EXACT lighting, time of day, white balance, and color grade \u2014 the same light as the ${carried} strip; no golden hour, no sunset, no warm relight, no cinematic regrade. Blend smoothly into the ${carried} strip with no visible seam. No people, no text, no labels, no watermarks.`;
+  }
+  return `${scene}This is a partial frame: the ${carried} portion contains real, finished pixels and the ${painted} portion is flat gray that MUST be painted in. Paint ONLY the ${painted} gray region as a natural, seamless continuation of the ${carried} portion \u2014 same scene, same perspective, continuing the horizon, geometry, and content across the boundary with no break. Keep the ${carried} portion completely unchanged. CRITICAL: do NOT change the lighting, exposure, white balance, or time of day. Match the ${carried} portion's EXACT light, color temperature, and contrast across the whole frame \u2014 if it is flat overcast daylight, keep flat overcast daylight. No golden hour, no sunset, no warm relight, no cinematic regrade. The seam between the ${carried} and ${painted} portions must be invisible. No people, no text, no labels, no watermarks.`;
+}
+
 exports.ACTION_FX = ACTION_FX;
 exports.ACTION_FX_CATEGORY_LABELS = ACTION_FX_CATEGORY_LABELS;
 exports.ACTION_FX_CATEGORY_ORDER = ACTION_FX_CATEGORY_ORDER;
@@ -32307,6 +32377,7 @@ exports.buildPostProcessHints = buildPostProcessHints;
 exports.buildReferenceBlocks = buildReferenceBlocks;
 exports.buildScenePrompt = buildScenePrompt;
 exports.buildStylingHints = buildStylingHints;
+exports.buildSurroundFillPrompt = buildSurroundFillPrompt;
 exports.buildTemporalHints = buildTemporalHints;
 exports.buildVoiceCharacterHints = buildVoiceCharacterHints;
 exports.buildVoiceDeliveryHints = buildVoiceDeliveryHints;
@@ -32449,6 +32520,7 @@ exports.getWardrobeEntry = getWardrobeEntry;
 exports.getWardrobePromptHint = getWardrobePromptHint;
 exports.groupFactoryPresets = groupFactoryPresets;
 exports.hasUpstreamCharacter = hasUpstreamCharacter;
+exports.identityRefsSentence = identityRefsSentence;
 exports.isAnalyzablePicker = isAnalyzablePicker;
 exports.isInstrumentalVocal = isInstrumentalVocal;
 exports.isVantageFraming = isVantageFraming;
@@ -32462,11 +32534,13 @@ exports.referenceRulesBlock = referenceRulesBlock;
 exports.renderStructuredFields = renderStructuredFields;
 exports.resolveBrandInput = resolveBrandInput;
 exports.resolveCharacterMentions = resolveCharacterMentions;
+exports.resolveGeminiOmniI2vInputs = resolveGeminiOmniI2vInputs;
 exports.resolveLocationMentions = resolveLocationMentions;
 exports.resolvePrompt = resolvePrompt;
 exports.resolveReferenceTokens = resolveReferenceTokens;
 exports.resolveSeedance2Inputs = resolveSeedance2Inputs;
 exports.resolveTemplate = resolveTemplate;
+exports.resolveVeoI2vInputs = resolveVeoI2vInputs;
 exports.resolveVideoReferenceCore = resolveVideoReferenceCore;
 exports.summarizePickerCatalogs = summarizePickerCatalogs;
 exports.toIdentityLockMode = toIdentityLockMode;