@slatesvideo/shared 0.5.1 → 0.5.3

package/dist/index.d.ts

@@ -5,4 +5,5 @@ export { SKILLS } from './skills/content.js';
 export * as operations from './operations/index.js';
 export { ALL_OPERATIONS, VIDEO_MODELS, defaultContext, type Operation, type OperationContext, type OperationResult } from './operations/index.js';
 export { MODEL_FACTS, getModelFact, type ModelFact } from './prompts/model-facts.js';
+export { PROMPTING_TIPS, getPromptingTips, type PromptingTipsEntry, type PromptingTipCard, type PromptingTipsKey } from './prompts/prompting-tips.js';
 //# sourceMappingURL=index.d.ts.map

package/dist/index.js

@@ -8,4 +8,7 @@ export { ALL_OPERATIONS, VIDEO_MODELS, defaultContext } from './operations/index
 // prompt derives its MODEL ROUTING doctrine from (kind: image vs video,
 // default/premium/niche notes). Edit model-facts.ts, never prose copies.
 export { MODEL_FACTS, getModelFact } from './prompts/model-facts.js';
+// Per-model prompting tips — the SSOT for the desktop "See prompting tips"
+// modals. The desktop renders these; it never hand-writes tips content.
+export { PROMPTING_TIPS, getPromptingTips } from './prompts/prompting-tips.js';
 //# sourceMappingURL=index.js.map

package/dist/operations/index.d.ts

@@ -67,6 +67,7 @@ export declare const uploadReferenceImage: Operation<{
     projectId: string;
     filePath?: string;
     dataUrl?: string;
+    type?: 'image' | 'video';
 }>;
 export declare const listFolders: Operation<{
     projectId: string;
@@ -169,7 +170,7 @@ export declare const editImage: Operation<{
     confirm?: boolean;
     background?: boolean;
 }>;
-export declare const VIDEO_MODELS: readonly ["kling-v3.0-std", "kling-v3.0-pro", "kling-v3.0-omni", "veo-3.1-fast", "veo-3.1-standard", "seedance-2"];
+export declare const VIDEO_MODELS: readonly ["kling-v3.0-std", "kling-v3.0-pro", "kling-v3.0-omni", "veo-3.1-fast", "veo-3.1-standard", "seedance-2", "omni-flash"];
 type VideoModel = (typeof VIDEO_MODELS)[number];
 export declare function videoCostKey(input: {
     model: VideoModel;
@@ -184,6 +185,7 @@ export declare function videoCostKey(input: {
     videoRefSeconds?: number;
 }): string;
 export declare function klingEditCostKey(model: 'kling-v3.0-omni-edit' | 'kling-v3.0-omni-pro-edit', duration: number): string;
+export declare function omniFlashEditCostKey(duration: number): string;
 export declare const generateVideo: Operation<{
     prompt: string;
     model: string;
@@ -255,13 +257,19 @@ export declare const editVideo: Operation<{
     projectId: string;
     sourceVideoAssetId: string;
     prompt: string;
-    model?: 'kling-v3.0-omni-edit' | 'kling-v3.0-omni-pro-edit';
+    model?: 'kling-v3.0-omni-edit' | 'kling-v3.0-omni-pro-edit' | 'omni-flash-edit';
     characterAssetIds?: string[];
     styleAssetIds?: string[];
     keepAudio?: boolean;
     background?: boolean;
     confirm?: boolean;
 }>;
+export declare const trimVideo: Operation<{
+    projectId: string;
+    assetId: string;
+    inSec?: number;
+    outSec: number;
+}>;
 export declare const getGenerationStatus: Operation<{
     generationId: string;
     waitSeconds?: number;
@@ -289,6 +297,30 @@ export declare const reorderClips: Operation<{
 export declare const removeClip: Operation<{
     clipId: string;
 }>;
+export declare const addTimelineTrack: Operation<{
+    projectId: string;
+    type?: 'video' | 'audio';
+    name?: string;
+}>;
+export declare const updateTimelineTrack: Operation<{
+    projectId: string;
+    trackId: string;
+    name?: string;
+    muted?: boolean;
+    locked?: boolean;
+    volume?: number;
+}>;
+export declare const removeTimelineTrack: Operation<{
+    projectId: string;
+    trackId: string;
+}>;
+export declare const updateTimelineSettings: Operation<{
+    projectId: string;
+    width?: number;
+    height?: number;
+    frameRate?: 24 | 30 | 60;
+    masterVolume?: number;
+}>;
 export declare const exportVideo: Operation<{
     projectId?: string;
     timelineId?: string;

package/dist/operations/index.js

@@ -439,12 +439,16 @@ export const getAssetVideoFrames = {
 };
 export const uploadReferenceImage = {
     id: 'slates_upload_reference_image',
-    description: 'Add a reference image to a Slates project. Pass either filePath (absolute path to a local file) or dataUrl (base64 data: URL) — exactly one.',
+    description: 'Add a reference image OR video clip to a Slates project. Pass either filePath (absolute path to a local file) or dataUrl (base64 data: URL) — exactly one. Set type:"video" on a filePath import to bring in a clip (the user\'s own footage to edit/relocate/trim); imported videos are probed on ingest, so duration + dimensions are available immediately. Default type is "image". dataUrl is image-only.',
     input: z
         .object({
         projectId: z.string().uuid(),
         filePath: z.string().optional(),
         dataUrl: z.string().optional(),
+        type: z
+            .enum(['image', 'video'])
+            .optional()
+            .describe('Asset kind for a filePath import — "image" (default) or "video". A dataUrl is always an image.'),
     })
         .refine((d) => !!d.filePath !== !!d.dataUrl, {
         message: 'Pass exactly one of filePath or dataUrl',
@@ -455,10 +459,13 @@ export const uploadReferenceImage = {
             const r = await desktop.post('/agent/assets/upload', {
                 projectId: input.projectId,
                 filePath: input.filePath,
-                type: 'image',
+                type: input.type ?? 'image',
             });
             return ok(r);
         }
+        if (input.type === 'video') {
+            throw new Error('dataUrl uploads are image-only — pass a filePath to import a video clip.');
+        }
         const r = await desktop.post('/agent/assets/upload-base64', {
             projectId: input.projectId,
             dataUrl: input.dataUrl,
@@ -1204,6 +1211,7 @@ export const VIDEO_MODELS = [
     'veo-3.1-fast',
     'veo-3.1-standard',
     'seedance-2',
+    'omni-flash',
 ];
 // Model → registry cost-key. Each provider's keys ship with their own
 // shape (verified against /api/agent/models):
@@ -1264,6 +1272,11 @@ export function videoCostKey(input) {
         }
         return `${tier}-${input.duration}s${input.sound === true ? '-audio' : ''}`;
     }
+    if (input.model === 'omni-flash') {
+        // Mirrors omniFlashCreditKey() in slate/src/shared/pricing.ts — flat 720p
+        // rate, audio native + included, no resolution/audio key dimension.
+        return `omni-flash-${input.duration}s`;
+    }
     throw new Error(`Unknown video model: ${input.model}`);
 }
 // Kling O3 video-to-video edit cost key — mirrors klingEditCreditKey() in
@@ -1274,6 +1287,12 @@ export function klingEditCostKey(model, duration) {
     const tier = model === 'kling-v3.0-omni-pro-edit' ? 'kling-v3-omni-pro-edit' : 'kling-v3-omni-edit';
     return `${tier}-${duration}s`;
 }
+// Omni Flash video-edit cost key — mirrors omniFlashCreditKey() in
+// slate/src/shared/pricing.ts (must byte-match; checked by the slates-api
+// pricing-consistency script). Duration is the CEILED source-clip length.
+export function omniFlashEditCostKey(duration) {
+    return `omni-flash-edit-${duration}s`;
+}
 /**
  * Forgiving model-id resolver. Agents routinely paste registry COST keys
  * ("kling-v3-standard-8s", "seedance-2-1080p-8s") into the `model` param —
@@ -1321,6 +1340,9 @@ function resolveVideoModel(raw) {
         seedance: 'seedance-2',
         'veo-3.1': 'veo-3.1-fast',
         'veo-3': 'veo-3.1-fast',
+        'gemini-omni-flash': 'omni-flash',
+        'gemini-omni-flash-preview': 'omni-flash',
+        'omni-flash-preview': 'omni-flash',
     };
     if (aliases[s]) {
         out.model = aliases[s];
@@ -1339,6 +1361,8 @@ function promptingSkillFor(model) {
         return 'slates-prompting-veo-3';
     if (model.startsWith('seedance'))
         return 'slates-prompting-seedance';
+    if (model.startsWith('omni-flash'))
+        return 'slates-prompting-omni-flash';
     return 'slates-cost-discipline';
 }
 export const generateVideo = {
@@ -1346,14 +1370,14 @@ export const generateVideo = {
     description: 'Generate video via Slates credits. REQUIRED before calling: read slates-model-selection (the routing doctrine), slates-cost-discipline, and the matching per-model prompting skill (slates-prompting-seedance / slates-prompting-kling-v3 / slates-prompting-veo-3) — video models prompt very differently; load them via slates_get_prompting_guide if no skill files are installed. Read slates-content-policy when the scene involves conflict, creatures, crowds, destruction, weapons, or young characters. projectId, aspectRatio, and duration are required (requires_clarification otherwise). Cost > $0.50 returns requires_confirm — pass confirm=true after explicit user OK. Image-to-video via firstFrameAssetId; first+last frames = Veo/Seedance only; ingredients via ingredientAssetIds (Kling Omni / Seedance). Asset params take UUIDs or badge codes ("IMG-A8").',
     input: z.object({
         prompt: z.string().min(1).max(4000),
-        model: z.string().describe('One of: kling-v3.0-std | kling-v3.0-pro | kling-v3.0-omni | seedance-2 | veo-3.1-fast | veo-3.1-standard. Pass the BASE id — duration and videoResolution are separate params (registry cost keys like "kling-v3-standard-8s" auto-resolve). Route per the slates-model-selection skill: Kling std = general-purpose DEFAULT, Seedance 2 = premium physics/effects/hero tier, Veo = native-synced-audio niche only (16:9, 4/6/8s) — never the default. All are VIDEO-only. For per-call cost, call slates_estimate_generation_cost — never quote prices from memory.'),
+        model: z.string().describe('One of: kling-v3.0-std | kling-v3.0-pro | kling-v3.0-omni | seedance-2 | veo-3.1-fast | veo-3.1-standard | omni-flash. Pass the BASE id — duration and videoResolution are separate params (registry cost keys like "kling-v3-standard-8s" auto-resolve). Route per the slates-model-selection skill: Kling std = general-purpose DEFAULT, Seedance 2 = premium physics/effects/hero tier, Veo = native-synced-audio niche only (16:9, 4/6/8s) — never the default, omni-flash = cheap 720p tier with audio included (3-10s, 16:9/9:16; t2v, single-start-frame i2v, or up to 7 reference images; no last frame / video / audio refs). All are VIDEO-only. For per-call cost, call slates_estimate_generation_cost — never quote prices from memory.'),
         projectId: z.string().uuid().optional().describe('Save into this Slates project. Strongly recommended — the desktop UI shows a progress card live and the asset appears when complete.'),
         aspectRatio: z.enum(['1:1', '16:9', '9:16', '4:3', '3:4', '21:9', '9:21', '4:5', '5:4', '2:3', '3:2']).optional().describe('Veo locks to 16:9 — passing anything else will be ignored or fail. Kling/Seedance support all.'),
-        duration: z.number().int().min(4).max(15).optional().describe('Seconds. Kling: 5-15. Veo: 4, 6, or 8 only (4K only at 8s). Seedance: 4-15. Default 5 if omitted but always be explicit (cost scales linearly).'),
+        duration: z.number().int().min(3).max(15).optional().describe('Seconds. Kling: 5-15. Veo: 4, 6, or 8 only (4K only at 8s). Seedance: 4-15. Omni Flash: 3-10. Default 5 if omitted but always be explicit (cost scales linearly).'),
         videoResolution: z.enum(['480p', '720p', '1080p', '4k']).optional().describe('Veo + Seedance. Seedance: 480p/720p/1080p/4K (default 1080p; 4K is native, the most expensive). Veo: 720p/1080p same price, 4K more (8s only).'),
         firstFrameAssetId: z.string().optional().describe('Starting frame for image-to-video: asset UUID or badge code ("IMG-A8") — codes resolve against the project at call time, so a code the user just spoke is always safe to pass.'),
         lastFrameAssetId: z.string().optional().describe('Ending frame (UUID or badge code). Veo and Seedance only. Pairs with firstFrameAssetId for guided transitions.'),
-        ingredientAssetIds: z.array(z.string()).max(9).optional().describe('Visual reference / ingredient assets (UUIDs or badge codes) for Kling Omni or Seedance. Up to 9 (Seedance) or 4 (Kling).'),
+        ingredientAssetIds: z.array(z.string()).max(9).optional().describe('Visual reference / ingredient assets (UUIDs or badge codes) for Kling Omni, Seedance, or Omni Flash. Up to 9 (Seedance), 4 (Kling), or 7 (Omni Flash, combined across all ref params).'),
         characterAssetIds: z.array(z.string()).optional().describe('Character sheet assets (UUIDs or badge codes) — keeps a character consistent across the shot.'),
         environmentAssetIds: z.array(z.string()).optional().describe('Environment grid assets (UUIDs or badge codes) — keeps a location/setting consistent across the shot.'),
         styleAssetIds: z.array(z.string()).optional().describe('Style reference assets (UUIDs or badge codes) — locks the visual style of the shot.'),
@@ -1412,10 +1436,41 @@ export const generateVideo = {
                 missing,
                 message: `Missing required field(s): ${missing.join(', ')}. ` +
                     `Read the slates-cost-discipline + ${promptingSkillFor(input.model)} skills, ` +
-                    `or ask the user. Veo locks to 16:9. Kling/Seedance support 1:1 16:9 9:16 4:3 3:4 21:9. ` +
-                    `Duration: Kling 5-15s, Veo 4/6/8s (4K only at 8s), Seedance 4-15s. Cost scales linearly with duration.`,
+                    `or ask the user. Veo locks to 16:9. Kling/Seedance support 1:1 16:9 9:16 4:3 3:4 21:9. Omni Flash: 16:9/9:16 only. ` +
+                    `Duration: Kling 5-15s, Veo 4/6/8s (4K only at 8s), Seedance 4-15s, Omni Flash 3-10s. Cost scales linearly with duration.`,
             });
         }
+        // Omni Flash: 3-10s, 720p only — t2v, single-start-frame i2v, or ref2v
+        // with up to 7 reference IMAGES. No last frame, no video/audio refs.
+        // Validate up front so the agent gets an actionable message instead of
+        // a registry throw.
+        if (input.model === 'omni-flash') {
+            if (input.duration < 3 || input.duration > 10) {
+                return ok({
+                    requires_clarification: true,
+                    missing: ['duration'],
+                    message: `Omni Flash supports 3-10 seconds (you passed ${input.duration}s). Pick a duration in that range.`,
+                });
+            }
+            if (input.lastFrameAssetId || input.videoReferenceAssetId || input.audioReferenceAssetId) {
+                return ok({
+                    requires_clarification: true,
+                    missing: [],
+                    message: 'Omni Flash takes a prompt, an optional start frame, and up to 7 reference IMAGES — last frames and video/audio references are not supported. Drop those, or switch to seedance-2 (video/audio refs, last frame) or veo (last frame).',
+                });
+            }
+            const refCount = (input.ingredientAssetIds?.length ?? 0) +
+                (input.characterAssetIds?.length ?? 0) +
+                (input.environmentAssetIds?.length ?? 0) +
+                (input.styleAssetIds?.length ?? 0);
+            if (refCount > 7) {
+                return ok({
+                    requires_clarification: true,
+                    missing: [],
+                    message: `Omni Flash takes at most 7 reference images combined (you passed ${refCount}). Trim the list.`,
+                });
+            }
+        }
         // Veo exists only at discrete durations 4/6/8s, and 4K only at 8s.
         // Validate up front so the agent gets an actionable message instead of
         // a generic "Model variant not in registry" throw from the cost lookup.
@@ -1503,7 +1558,7 @@ export const generateVideo = {
         const entry = registry.models.find((m) => m.model === costKey);
         if (!entry) {
             throw new Error(`Model variant not in registry: ${costKey}. ` +
-                `Available video models: ${registry.models.filter((m) => m.model.startsWith('kling') || m.model.startsWith('veo') || m.model.startsWith('seedance')).map((m) => m.model).slice(0, 20).join(', ')}`);
+                `Available video models: ${registry.models.filter((m) => m.model.startsWith('kling') || m.model.startsWith('veo') || m.model.startsWith('seedance') || m.model.startsWith('omni-flash')).map((m) => m.model).slice(0, 20).join(', ')}`);
         }
         const totalCents = creditCost(entry);
         // Pre-flight confirm gate. Fires when:
@@ -1955,15 +2010,15 @@ export const generateMotionTransfer = {
 // ── Edit video (Kling O3 video-to-video) ────────────────────────
 export const editVideo = {
     id: 'slates_edit_video',
-    description: 'Edit an EXISTING video clip with one instruction via Kling O3 video-to-video edit — character swap, environment change, style transfer — in one pass, no masking. Original motion, camera, and audio are preserved; only what the prompt names changes. Use when a clip is ~90% right (fix it, don\'t re-roll it) or to AI-edit the user\'s own footage. Source clip constraints: 3–15s, 720–3840px, MP4/MOV. Cost = per second of OUTPUT (≈ clip length, rounded UP to the next second): kling-v3.0-omni-edit ≈ 19¢/s, kling-v3.0-omni-pro-edit ≈ 25¢/s. Subjects to swap IN go as characterAssetIds (frontal + angle images become Kling elements); style refs as styleAssetIds; max 4 combined. The edited clip saves as a NEW asset linked to its parent (chain edits freely). Routing: Kling edit is the default edit tool (element lock + audio intact); prefer Seedance edit/relocate only for style-transfer-heavy jobs — see slates-model-selection. Prompting: slates-prompting-kling-v3 §Edit.',
+    description: 'Edit an EXISTING video clip with one instruction — character swap, environment change, style transfer — in one pass, no masking. Original motion, camera, and audio are preserved; only what the prompt names changes. Use when a clip is ~90% right (fix it, don\'t re-roll it) or to AI-edit the user\'s own footage. Engines: Kling O3 edit (default; 3–15s clips, 720–3840px, subject/style refs via elements) or omni-flash-edit (Gemini Omni Flash; 3–10s clips, 720p output, PROMPT-ONLY — no refs, cheapest seat). Cost = per second of OUTPUT (≈ clip length, rounded UP to the next second): omni-flash-edit ≈ 19¢/s ≈ kling-v3.0-omni-edit ≈ 19¢/s, kling-v3.0-omni-pro-edit ≈ 25¢/s. Subjects to swap IN go as characterAssetIds (frontal + angle images become Kling elements — Kling models only); style refs as styleAssetIds; max 4 combined. The edited clip saves as a NEW asset linked to its parent (chain edits freely). Routing: Kling edit is the default edit tool (element lock + audio intact); omni-flash-edit for cheap prompt-only footage-synced swaps; prefer Seedance edit/relocate only for style-transfer-heavy jobs — see slates-model-selection. Prompting: slates-prompting-kling-v3 §Edit / slates-prompting-omni-flash.',
     input: z.object({
         projectId: z.string().uuid().describe('Project the source clip lives in.'),
-        sourceVideoAssetId: z.string().describe('The VIDEO asset to edit — UUID or badge code ("VID-V3", bare "V3"); codes resolve against the project at call time. 3–15s clips only.'),
-        prompt: z.string().min(1).max(2500).describe('The change, not the whole scene — e.g. "replace the man with @marcus", "make it a rainy night", "turn the street into a neon Tokyo alley". Mention subjects with @name; the transport compiles them to Kling\'s @ElementN notation.'),
-        model: z.enum(['kling-v3.0-omni-edit', 'kling-v3.0-omni-pro-edit']).optional().describe('Default kling-v3.0-omni-edit. Pro (~25¢/s vs ~19¢/s) only for hero shots where fidelity matters.'),
-        characterAssetIds: z.array(z.string()).max(4).optional().describe('Subject/element image assets to swap IN (UUIDs or badge codes). Each becomes a Kling element (@ElementN).'),
-        styleAssetIds: z.array(z.string()).max(4).optional().describe('Style/appearance reference images (@ImageN). Max 4 combined with characterAssetIds.'),
-        keepAudio: z.boolean().optional().describe('Preserve the original audio track (default true).'),
+        sourceVideoAssetId: z.string().describe('The VIDEO asset to edit — UUID or badge code ("VID-V3", bare "V3"); codes resolve against the project at call time. Kling: 3–15s clips; omni-flash-edit: 3–10s.'),
+        prompt: z.string().min(1).max(2500).describe('The change, not the whole scene — e.g. "replace the man with @marcus", "make it a rainy night", "turn the street into a neon Tokyo alley". Mention subjects with @name; the transport compiles them to Kling\'s @ElementN notation (Kling models). For omni-flash-edit keep it simple and add "Keep everything else the same."'),
+        model: z.enum(['kling-v3.0-omni-edit', 'kling-v3.0-omni-pro-edit', 'omni-flash-edit']).optional().describe('Default kling-v3.0-omni-edit. Pro (~25¢/s vs ~19¢/s) only for hero shots where fidelity matters. omni-flash-edit (~19¢/s, 720p, 3–10s) for prompt-only edits — it takes NO character/style refs.'),
+        characterAssetIds: z.array(z.string()).max(4).optional().describe('Subject/element image assets to swap IN (UUIDs or badge codes). Each becomes a Kling element (@ElementN). KLING MODELS ONLY — rejected on omni-flash-edit.'),
+        styleAssetIds: z.array(z.string()).max(4).optional().describe('Style/appearance reference images (@ImageN). Max 4 combined with characterAssetIds. KLING MODELS ONLY — rejected on omni-flash-edit.'),
+        keepAudio: z.boolean().optional().describe('Preserve the original audio track (default true; Kling models only — omni-flash-edit output carries its own audio).'),
         background: z.boolean().optional().describe(BACKGROUND_DESCRIBE),
         confirm: z.boolean().optional().describe('Set true to bypass the cost confirm gate after the user OKs the spend.'),
     }),
@@ -1974,6 +2029,12 @@ export const editVideo = {
             await desktop.requireCapability('background-generation', 'background generation');
         }
         const model = input.model ?? 'kling-v3.0-omni-edit';
+        const isOmniFlashEdit = model === 'omni-flash-edit';
+        // Kling edit: 3–15s source clips; Omni Flash edit: 3–10s.
+        const maxClipSeconds = isOmniFlashEdit ? 10 : 15;
+        if (isOmniFlashEdit && ((input.characterAssetIds?.length ?? 0) > 0 || (input.styleAssetIds?.length ?? 0) > 0)) {
+            throw new Error('omni-flash-edit is prompt-only — it takes no character/style reference images. Drop the refs, or switch to kling-v3.0-omni-edit which supports elements.');
+        }
         // Resolve refs (UUIDs or badge codes) against the project AT CALL TIME.
         const refInputs = [
             { ref: input.sourceVideoAssetId, role: 'source clip' },
@@ -2001,11 +2062,12 @@ export const editVideo = {
         if (!Number.isFinite(clipSeconds) || clipSeconds <= 0) {
             throw new Error('Source clip has no recorded duration — cannot quote the edit. Re-import the clip or pick another.');
         }
-        if (clipSeconds > 15.05 || clipSeconds < 2.95) {
-            throw new Error(`Source clip is ${clipSeconds.toFixed(1)}s — Kling O3 edit accepts 3–15s. Trim it first (agents can pre-trim on the timeline).`);
+        if (clipSeconds > maxClipSeconds + 0.05 || clipSeconds < 2.95) {
+            throw new Error(`Source clip is ${clipSeconds.toFixed(1)}s — ${model} accepts 3–${maxClipSeconds}s. ` +
+                `Trim it first with slates_trim_video (e.g. inSec 0, outSec ${maxClipSeconds}), then edit the trimmed clip.`);
         }
-        const billedSeconds = Math.min(15, Math.max(3, Math.ceil(clipSeconds - 0.05)));
-        const costKey = klingEditCostKey(model, billedSeconds);
+        const billedSeconds = Math.min(maxClipSeconds, Math.max(3, Math.ceil(clipSeconds - 0.05)));
+        const costKey = isOmniFlashEdit ? omniFlashEditCostKey(billedSeconds) : klingEditCostKey(model, billedSeconds);
         const cloud = ctx.cloud();
         const registry = await cloud.get('/api/agent/models');
         const entry = registry.models.find((m) => m.model === costKey);
@@ -2078,6 +2140,41 @@ export const editVideo = {
         };
     },
 };
+// ── Trim a video to an exact window (fit-to-model primitive) ────
+export const trimVideo = {
+    id: 'slates_trim_video',
+    description: 'Trim a video asset to an exact [inSec, outSec] window and save the result as a NEW clip linked to the original (the original is untouched). This is the fit-to-model primitive: an 11s clip will not run on omni-flash-edit (3–10s) or Kling edit (3–15s), and a Seedance video reference must be 2–15s — trim it first, then edit/relocate the trimmed clip. EXACT re-encode (not a keyframe-snapped cut) so the result honors hard duration caps to the frame; any phone rotation flag is baked into the pixels in the same pass. The new clip lands with correct duration/width/height immediately. inSec defaults to 0.',
+    input: z.object({
+        projectId: z.string().uuid().describe('Project the clip lives in.'),
+        assetId: z
+            .string()
+            .describe('The VIDEO asset to trim — UUID or badge code ("VID-V3", bare "V3"); resolves against the project at call time.'),
+        inSec: z.number().min(0).optional().describe('Trim start in seconds (default 0).'),
+        outSec: z.number().positive().describe('Trim end in seconds. Must be greater than inSec.'),
+    }),
+    async run(input, ctx) {
+        const desktop = ctx.desktop();
+        const resolved = await resolveAssetRefs(ctx, input.projectId, [input.assetId]);
+        const assetId = resolved.get(input.assetId)?.id ?? input.assetId;
+        const inSec = input.inSec ?? 0;
+        if (input.outSec - inSec < 0.05) {
+            throw new Error('outSec must be at least ~0.1s after inSec.');
+        }
+        const r = await desktop.post('/agent/assets/trim-video', {
+            assetId,
+            inSec,
+            outSec: input.outSec,
+        });
+        const a = r.asset;
+        const name = a?.code ?? a?.id ?? 'new clip';
+        return {
+            text: `Trimmed clip saved as ${name}${a?.label ? ` — ${a.label}` : ''} ` +
+                `(${(input.outSec - inSec).toFixed(1)}s, ${inSec.toFixed(1)}–${input.outSec.toFixed(1)}s). ` +
+                `Edit or generate from it by its new id/code.`,
+            data: { asset: r.asset },
+        };
+    },
+};
 // ── Generation status (background mode) ─────────────────────────
 export const getGenerationStatus = {
     id: 'slates_get_generation_status',
@@ -2163,11 +2260,11 @@ export const getTimeline = {
 };
 export const addClipToTimeline = {
     id: 'slates_add_clip_to_timeline',
-    description: "Append a video asset from the project to the project's timeline (or place it at an explicit startFrame). Defaults match the desktop UI: clip goes to the end of the first video track, full source duration, and an empty timeline auto-adopts the clip's resolution and frame rate. Optionally trim with sourceInFrame/sourceOutFrame (frames at the SOURCE fps). Use slates_get_timeline first to see current clips and pick positions. Only video assets are accepted.",
+    description: "Append a video or audio asset from the project to the project's timeline (or place it at an explicit startFrame). Defaults match the desktop UI: video clips go to the end of the first video track; audio clips (music, voiceover, AI audio) go after the last clip on the first AUDIO track and are mixed under the video on export. An empty timeline auto-adopts the first video clip's resolution and frame rate; later higher-resolution clips raise the canvas. Overlapping video clips resolve top-track-wins. Optionally trim with sourceInFrame/sourceOutFrame (frames at the SOURCE fps). Use slates_get_timeline first to see current clips and pick positions.",
     input: z.object({
         projectId: z.string().uuid(),
-        assetId: z.string().uuid().describe('Video asset already in the project.'),
-        trackId: z.string().uuid().optional().describe('Target track. Default: the first video track.'),
+        assetId: z.string().uuid().describe('Video or audio asset already in the project.'),
+        trackId: z.string().uuid().optional().describe('Target track (type must match the asset: video asset → video track, audio asset → audio track). Default: the first track of the matching type.'),
         startFrame: z.number().int().min(0).optional().describe('Timeline frame to place the clip at. Default: append after the last clip.'),
         sourceInFrame: z.number().int().min(0).optional(),
         sourceOutFrame: z.number().int().min(1).optional(),
@@ -2218,6 +2315,66 @@ export const removeClip = {
         return ok(await desktop.post('/agent/timeline/remove-clip', input));
     },
 };
+export const addTimelineTrack = {
+    id: 'slates_add_timeline_track',
+    description: "Add a track to the project's timeline (default: an audio track, for layering voiceover + music + AI audio). The new track is appended below existing tracks. Returns the new track and the full timeline.",
+    input: z.object({
+        projectId: z.string().uuid(),
+        type: z.enum(['video', 'audio']).optional().describe('Default: audio.'),
+        name: z.string().optional().describe("Default: 'Audio N' / 'Video N'."),
+    }),
+    async run(input, ctx) {
+        const desktop = ctx.desktop();
+        await desktop.requireCapability('timeline-tracks', 'timeline tracks + audio mixing');
+        return ok(await desktop.post('/agent/timeline/add-track', input));
+    },
+};
+export const updateTimelineTrack = {
+    id: 'slates_update_timeline_track',
+    description: 'Update a timeline track: rename, mute/unmute, lock/unlock, or set its volume fader (linear gain, -∞ to +12 dB). Track volume applies to both preview and MP4 export — audio-track clips are mixed at this gain; a muted video track still shows video but its embedded audio is silenced.',
+    input: z.object({
+        projectId: z.string().uuid(),
+        trackId: z.string().uuid(),
+        name: z.string().optional(),
+        muted: z.boolean().optional(),
+        locked: z.boolean().optional(),
+        volume: z.number().min(0).max(4).optional().describe('Track fader as LINEAR gain: 0 = -∞ (silent), 1 = 0 dB (unity), ~3.98 = +12 dB (max boost).'),
+    }),
+    async run(input, ctx) {
+        const desktop = ctx.desktop();
+        await desktop.requireCapability('timeline-tracks', 'timeline tracks + audio mixing');
+        return ok(await desktop.post('/agent/timeline/update-track', input));
+    },
+};
+export const removeTimelineTrack = {
+    id: 'slates_remove_timeline_track',
+    description: 'Remove an EMPTY timeline track (fails if it still has clips, or if it is the last track of its type).',
+    input: z.object({
+        projectId: z.string().uuid(),
+        trackId: z.string().uuid(),
+    }),
+    async run(input, ctx) {
+        const desktop = ctx.desktop();
+        await desktop.requireCapability('timeline-tracks', 'timeline tracks + audio mixing');
+        return ok(await desktop.post('/agent/timeline/remove-track', input));
+    },
+};
+export const updateTimelineSettings = {
+    id: 'slates_update_timeline_settings',
+    description: "Update the project timeline's output settings: resolution, frame rate (24/30/60 — all clips are conformed to it on export), and masterVolume, the output fader (linear gain, -∞ to +12 dB) applied to the final mix in both preview and MP4 export (use it to prevent clipping when stacking loud tracks). Note these are normally auto-managed: the first video clip sets fps + resolution, and higher-res clips raise the canvas. Changing frameRate after clips are placed retimes them — avoid unless the timeline is empty.",
+    input: z.object({
+        projectId: z.string().uuid(),
+        width: z.number().int().min(16).optional(),
+        height: z.number().int().min(16).optional(),
+        frameRate: z.union([z.literal(24), z.literal(30), z.literal(60)]).optional(),
+        masterVolume: z.number().min(0).max(4).optional().describe('Output fader as LINEAR gain: 0 = -∞ (silent), 1 = 0 dB (unity), ~3.98 = +12 dB (max boost).'),
+    }),
+    async run(input, ctx) {
+        const desktop = ctx.desktop();
+        await desktop.requireCapability('timeline-tracks', 'timeline tracks + audio mixing');
+        return ok(await desktop.post('/agent/timeline/update-settings', input));
+    },
+};
 // ── Export ──────────────────────────────────────────────────────
 const ABSOLUTE_PATH_RE = /^([A-Za-z]:[\\/]|\/)/;
 export const exportVideo = {
@@ -2584,6 +2741,8 @@ function resolveGuideTopic(topic) {
         return 'slates-prompting-seedream-5-lite';
     if (t.startsWith('veo'))
         return 'slates-prompting-veo-3';
+    if (t.startsWith('omni-flash') || t.startsWith('gemini-omni') || t === 'omni flash')
+        return 'slates-prompting-omni-flash';
     if (t.startsWith('kling-mc'))
         return 'slates-prompting-motion-transfer';
     if (t === 'edit-video' || t === 'video-edit' || t === 'edit video' || t === 'video edit')
@@ -2666,6 +2825,7 @@ export const ALL_OPERATIONS = [
     generateLipSync,
     generateMotionTransfer,
     editVideo,
+    trimVideo,
     editImage,
     getGenerationStatus,
     listGenerations,
@@ -2673,6 +2833,10 @@ export const ALL_OPERATIONS = [
     addClipToTimeline,
     reorderClips,
     removeClip,
+    addTimelineTrack,
+    updateTimelineTrack,
+    removeTimelineTrack,
+    updateTimelineSettings,
     exportVideo,
     exportTimelineXml,
     revealFile,

package/dist/prompts/index.d.ts

@@ -5,4 +5,5 @@ export * from './style-library.js';
 export * from './model-facts.js';
 export * from './character-sheet.js';
 export * from './environment-sheet.js';
+export * from './prompting-tips.js';
 //# sourceMappingURL=index.d.ts.map

package/dist/prompts/index.js

@@ -16,4 +16,8 @@ export * from './style-library.js';
 export * from './model-facts.js';
 export * from './character-sheet.js';
 export * from './environment-sheet.js';
+// Exported from the `./prompts` subpath (not just the root barrel) because the
+// desktop RENDERER imports the tips — the root barrel re-exports auth.js
+// (node:fs/os/path), which breaks browser bundling. `./prompts` stays Node-free.
+export * from './prompting-tips.js';
 //# sourceMappingURL=index.js.map

package/dist/prompts/model-facts.js

@@ -58,7 +58,7 @@ export const MODEL_FACTS = [
         kind: 'video',
         maxRefImages: null,
         maxIngredients: 9, // ingredient images per video gen
-        notes: 'PREMIUM video tier — route here the moment physics, effects, destruction, or scale matter, and for hero shots. VIDEO-ONLY: cannot generate standalone images (use NB2/FLUX.2/Seedream for those). Up to 9 ingredient images. Strong I2V / own-footage restyle. Native 4K. Also the PREMIUM engine inside the Motion Transfer and Lip Sync tools (single-pass: driving video / dialogue are native conditioning signals — better motion fidelity, natural speech, voice cloned from a video source; video references bill input+output seconds).',
+        notes: 'PREMIUM video tier — route here the moment physics, effects, destruction, or scale matter, and for hero shots. VIDEO-ONLY: cannot generate standalone images (use NB2/FLUX.2/Seedream for those). Up to 9 ingredient images. Strong I2V / own-footage restyle. Native 4K, but 4K VIDEO is a Pro-only tier gate (base maxes at 1080p; server returns PRO_REQUIRED) — default 1080p unless the user is on Pro. Also the PREMIUM engine inside the Motion Transfer and Lip Sync tools (single-pass: driving video / dialogue are native conditioning signals — better motion fidelity, natural speech, voice cloned from a video source; video references bill input+output seconds).',
     },
     {
         id: 'kling-v3',
@@ -74,7 +74,7 @@ export const MODEL_FACTS = [
         kind: 'video',
         maxRefImages: null,
         maxIngredients: 4, // combined subject elements + style refs per edit
-        notes: 'VIDEO-TO-VIDEO EDIT (default edit tool) — takes an EXISTING 3–15s clip and changes only what the prompt names: character swap, environment change, style transfer. Original motion/camera/audio preserved. One pass, no masking. Use to FIX a 90%-right clip instead of re-rolling it, or to AI-edit the user\'s own footage. Elements (@ElementN = frontal + angles) lock subject identity; max 4 combined refs. Billed per second of output (≈ clip length, rounded up). Seedance edit/relocate is the alternative for style-transfer-heavy jobs.',
+        notes: 'VIDEO-TO-VIDEO EDIT — the REF-DRIVEN edit tool: takes an EXISTING 3–15s clip and changes what the prompt names, with element/style reference images (@ElementN = frontal + angles) locking subject identity; max 4 combined refs. keep_audio preserves the ORIGINAL audio verbatim (spoken words cannot drift) — but video lips can drift slightly against it, and multi-beat instructions get under-executed (7/09 receipt: missed a second action beat Omni Flash edit landed) — ONE beat per pass. Route here when an edit NEEDS reference images or bit-exact audio; for prompt-only footage-synced VFX, omni-flash-edit won the 7/09 fidelity head-to-head. Billed per second of output (≈ clip length, rounded up). Seedance edit/relocate is the alternative for style-transfer-heavy jobs.',
     },
     {
         id: 'veo-3.1',
@@ -84,6 +84,22 @@ export const MODEL_FACTS = [
         maxIngredients: 3,
         notes: 'NICHE, never the default — pick only when native synchronized audio must generate WITH the video in one gen. 16:9 only, 4/6/8s only. Otherwise Kling (default) or Seedance (physics/premium) win.',
     },
+    {
+        id: 'omni-flash',
+        label: 'Gemini Omni Flash',
+        kind: 'video',
+        maxRefImages: null,
+        maxIngredients: 7, // ref2v image_urls; 7 mirrors Google's own reference limit
+        notes: 'CHEAP 720p tier with native synced audio included — t2v, single-start-frame i2v, or reference-to-video with up to 7 reference images. 3-10s, 16:9/9:16 only. No last frame, no video/audio references. VIDEO-ONLY. New seat: quality vs Kling/Seedance unproven pending comparison gens — do not route hero shots here; use it for cheap drafts, audio-in-one-gen at low cost, ref2v character consistency trials, and its edit variant.',
+    },
+    {
+        id: 'omni-flash-edit',
+        label: 'Omni Flash Edit',
+        kind: 'video',
+        maxRefImages: null,
+        maxIngredients: 0, // prompt + source clip ONLY — no element/style refs on this endpoint
+        notes: 'VIDEO-TO-VIDEO EDIT, prompt-only — THE EDIT-FIDELITY WINNER (7/09 head-to-head vs Kling edit on real talking footage: lips held perfectly, audio near-identical, both action beats landed). Takes an EXISTING 3-10s clip and changes what the prompt names, footage-synced (prop/effect/environment/lighting swaps). Fidelity is EARNED by prompt discipline: ONE short instruction + "Keep everything else the same." — long descriptive prompts DESTROY it (Google-documented + 7/09 receipt). Never name objects as metaphors ("candle-like" → literal candle). Quirk: occasional tail jitter/doubled last speech beat — trim the tail. NO reference images (identity swaps needing refs → Kling edit); bit-exact audio needs → Kling keep_audio or segment-splice. 720p output, cheapest edit seat (~2/3 of Kling edit Std).',
+    },
 ];
 const FACT_BY_ID = new Map(MODEL_FACTS.map((m) => [m.id, m]));
 export function getModelFact(id) {

package/dist/prompts/prompting-tips.d.ts

@@ -0,0 +1,23 @@
+export interface PromptingTipCard {
+    heading: string;
+    /** Monospace example line(s). \n renders as a line break. */
+    example?: string;
+    note: string;
+    /** Render highlighted as a critical/mandatory card. */
+    critical?: boolean;
+}
+export interface PromptingTipsEntry {
+    /** Family label for the modal title ("Prompting Tips — {label}"). */
+    label: string;
+    /** Intro paragraphs above the cards. */
+    intro: string[];
+    /** Two columns of tip cards. */
+    columns: [PromptingTipCard[], PromptingTipCard[]];
+    /** Footer callout paragraphs. */
+    footer?: string[];
+}
+export type PromptingTipsKey = 'seedance' | 'kling' | 'kling-edit' | 'veo' | 'omni-flash' | 'omni-flash-edit' | 'nano-banana' | 'nano-banana-lite';
+export declare const PROMPTING_TIPS: Record<PromptingTipsKey, PromptingTipsEntry>;
+/** Null when no tips exist for the key — callers render an honest fallback. */
+export declare function getPromptingTips(key: string): PromptingTipsEntry | null;
+//# sourceMappingURL=prompting-tips.d.ts.map