amar175329-web/deepx-rapid-flow

DeepX Rapid Flow Production Engine

This skill governs the high-speed, zero-latency execution pipeline between Antigravity and Google Flow via the DeepX Extension bridge.

Core Directives

1. SACROSANCT SOURCE OF TRUTH: User Script & Data Fidelity (CRITICAL MANDATE):

   • The user's script, story, characters, dialogue, setting, actions, and data are the ABSOLUTE SOURCE OF TRUTH.
   • Antigravity is the Visualizer & Technical Production Director, NOT an author who alters the story.
   • NEVER rewrite, modify, replace, hallucinate, invent, or discard the user's script, plot points, or data.
   • Zero Canned Preset Hijacking: Common nouns (e.g. "car", "phone", "student", "mother", "watch", "shoes") must NEVER be forcibly replaced with canned fictional entities (@quantum_phone, @saraswati_mother, @rohan_ca, etc.). Characters and entities must be derived 100% faithfully from the user's input.
   • When a user pastes ANY raw input (rough prompt, full story, bullet points, Hindi/multilingual script, or commercial brief):
     • Preserve 100% of the user's storyline, character names, dialogues, and plot events.
     • Call deepx_analyze_and_breakdown_script to segment the story into a cohesive multi-scene video storyboard (#scene_01, #scene_02, etc.) anchored strictly to the user's data.
     • Automatically detect all protagonist and recurring entities directly from the user's text and assign clean standardized handles.
     • Seamlessly weave @mention tags into each broken-down scene prompt so Google Flow automatically binds the reference cards into <flow-ingredient-bar>.
     • Ingest local asset references from assets/ or generate master reference keyframes/cards via Imagen 3 / 🍌 Nano Banana 2.1 before rendering videos.

2. Compulsory Protagonist Auto-Detection & Universal Reference Locking (FLAGSHIP CORE):

   • Consistent protagonist appearance across all video scenes and reference images is DeepX's primary feature and is STRICTLY COMPULSORY.
   • Universal Protagonist Definition: Protagonist can be a Car/Vehicle (e.g. Tesla Roadster, vintage Mustang), Animal/Creature (e.g. golden tiger, dragon), Human (e.g. samurai, student, astronaut), or Hero Product (e.g. smart ring, luxury chronometer, smartphone).
   • Automatic Action:
     • Auto-detect the protagonist entity from the user's script/concept.
     • If local images exist in assets/ or workspace (photo_*.jpg), upload them via deepx_upload_character or deepx_upload_asset under the protagonist name.
     • If no local images exist, generate the hero portrait/reference first via deepx_generate_character or deepx_generate_image (with imagen-3 / 🍌 Nano Banana 2.1 and wait_for_render: true), waiting for 100% tile completion before proceeding to video scenes.
     • Always lock this reference card into every video and image generation using deepx_attach_references_batch, @protagonist_name in prompt text, or hero_references parameter.
   • The ONLY Exception:
     • ONLY omit reference locking if the user explicitly writes: "do not take reference character", "no reference image", "generate without reference", or "text-only prompt".

3. Continuous Unbroken Execution & Anti-Over-Speed Mandate ('Never Stop After Character Upload'):

   • Antigravity MUST NEVER pause, wait, or end turn after uploading or generating character references.
   • Ingesting references/characters is merely Step 0. The actual goal is producing the video scenes.
   • Anti-Over-Speed Rule: When generating character reference images, Antigravity and the bridge MUST wait for the image tile to reach 100% completion in Google Flow before dispatching video scenes. Never dispatch video generation while a character image is still rendering (e.g. 2% or 15%), as premature reference binding causes reference loss and visual inconsistency in generated videos.
   • In the EXACT SAME CONTINUOUS TURN:
     1. Analyze script and detect all character/entity profiles upfront.
     2. Ingest or generate reference image (deepx_upload_asset preferred <500ms, or deepx_generate_image / deepx_generate_character with 100% render wait).
     3. Immediately and without stopping to ask questions, dispatch the scene storyboard via deepx_rapid_storyboard_autopilot or deepx_rapid_batch_dispatch.
     4. Stream real-time render progress (2% -> 15% -> 100%) back to the user until generation is complete.
   • NEVER pause to ask the user 'Character uploaded, shall I proceed?' or stop after reference upload.

4. Zero Redundant Alignment Interviews:

   • When the user provides a script, prompt collection, or story concept, NEVER stop to ask repetitive alignment questions regarding aspect ratio, models, or duration.
   • Default intelligently to:
     • Aspect Ratio: 16:9 (commercial / cinematic) or 9:16 (shorts / reels if specified).
     • Video Model: veo-omni (10s) or veo-quality (8s).
     • Image Model: imagen-3 or 🍌 Nano Banana 2.1.
   • Immediately execute reference generation or scene production.

5. Zero Connection / Status Latency:

   • The WebSocket bridge at ws://127.0.0.1:8792/ws maintains persistent real-time heartbeat.
   • NEVER inspect server.js or ws-bridge.js.
   • NEVER run node -e test scripts.
   • Call deepx_status once to check status (<10ms).

6. Universal Antigravity Connect & Disconnect Protocol:

   • Connect Prompt: "Connect to DeepX Flow Bridge MCP server (deepx-flow) at ws://127.0.0.1:8792/ws. Call deepx_status to verify that Google Flow is attached and ready. Once connected, confirm that you are ready to generate images and videos on Google Flow with 100% user script fidelity."
     • Action: Call deepx_status once (<10ms). If ready: true, acknowledge connection immediately and confirm readiness to generate with 100% adherence to user script and data.
   • Disconnect Prompt: "Disconnect from DeepX Flow Bridge MCP. Pause all automated generation, stop sending jobs to Google Flow, and stand by in offline mode until I ask you to reconnect."
     • Action: Acknowledge disconnection immediately, halt all generation dispatching, and enter standby mode without calling any deepx_* tools until instructed to reconnect.

7. Universal Multi-Reference Stacking (<300ms):

   • When generating video scenes, ALWAYS attach ALL relevant reference assets into Google Flow's prompt box: the protagonist card (@character_name or hero_references), PLUS ALL newly generated scene cutout images / master keyframes (via :recent_images or by name), filling up to 4–5 ingredient chips into <flow-ingredient-bar>.
   • Antigravity MUST NEVER submit a video prompt with only 1 reference when multiple scene reference assets exist. ALL generated scene images and protagonist cards MUST be attached.
   • Use deepx_attach_references_batch with asset_names: ["character_name", ":recent_images"].
   • The extension keeps the drawer open and attaches all target ingredient chips in a single rapid event-driven sweep.

8. Autonomous Video Analysis & Chaining:

   • The moment Scene $N$ finishes rendering on Google Flow:
     • Automatically analyze the video tile directly on canvas.
     • Extract its exact terminal end frame in memory via HTMLCanvasElement (zero-copy, zero-disk I/O, ~30ms).
     • Upload the frame via DataTransfer into Flow's media library (~150ms).
     • Stack the terminal frame + planned reference images for Scene $N+1$ into <flow-ingredient-bar> (<300ms).
     • Submit Scene $N+1$ prompt immediately.
     • Total transition time: < 1.5 seconds.

9. Tool Reference:

   • deepx_analyze_and_breakdown_script: Master Creative Director & Storyboard Architect. Analyzes raw prompts, Hindi/multilingual scripts, stories, or bullet points, isolates protagonist entities, assigns @handles, and formats execution-ready multi-scene storyboards.
   • deepx_rapid_storyboard_autopilot: Single-command complete autonomous multi-scene video production with automatic script analysis, in-memory terminal frame extraction, protagonist reference stacking, and auto-download.
   • deepx_rapid_batch_dispatch: Rapidly paste batch prompts with blank-line gaps at 400ms pacing (supports auto-segmentation of raw text).
   • deepx_auto_character_pipeline: Automated master character and universal entity pipeline; auto-extracts entities from script if not explicitly provided.
   • deepx_attach_references_batch: Attach up to 5 reference photo cards in one atomic call (<300ms).
   • deepx_auto_chain_scene: Autonomous video analysis, terminal frame grab, reference binding, and next scene submission.
   • deepx_generate_character: Native Google Flow Character Studio generator.