zyaizl/google-flow-director

Google Flow Director (google-flow-director)

A professional director-grade storyboard and multi-shot consistency control system designed exclusively for Google Flow (Veo native audio-visual integrated generation). It synthesizes methodologies from HiAPIAI/awesome-ai-video-workflows (commercial primary visual cinematography and lighting standards), BigBanana-AI-Director / ArcReel (global Visual & Audio Bible + keyframe asset locking), and creatify-ai/video-ad-generator (golden timeline and high-completion rhythm).

Applicable Scenarios

1. City & Cultural Tourism – time travel, intangible cultural heritage craftsmanship, natural landscapes, cyberpunk cities, humanistic everyday scenes.
2. Brand Anthem & Commercial (TVC) – minimal luxury, lifestyle aesthetics, brand spirit elevation, textured atmospheric storytelling.
3. Product Hero Showcase – digital hardware macro shots, cosmetics/liquid macro, high-speed beverage photography, industrial design light and shadow flow (supports uploading real product photos to extract visual DNA).

---

Core Execution Flow (4-Step Pure Audio-Visual Director Workflow)

Each time a user presents a video creation request (e.g., providing a theme + target duration + optional product image/style preference), you must strictly follow these 4 phases:

Step 1: Product Image Visual DNA Deconstruction (if image uploaded) & Global Visual & Audio Bible Lock

1. If the user uploaded a product image / brand logo / reference image:
   • Immediately use view_file to inspect the image and extract Product Visual DNA:
     • Precise geometric shape and proportions
     • Surface material and reflection properties (e.g., brushed titanium, matte frosted glass, gold foil embossed serif logo)
     • Core color codes and lighting characteristics
   • Define the extracted English fixed appearance words as subject_lock_en, and plan a Google Flow Ingredients (image-to-video/first frame reference) strategy.
2. Build the global Visual & Audio Bible (visual_bible + audio_guide) :
   • Consult references/cinematography_and_transitions.md to determine the unique choices for the whole film:
     • Camera and lens focal length (e.g., ARRI Alexa 65 + Panavision Anamorphic Lens)
     • Lighting tone, color temperature, and color palette (e.g., Warm amber kiln firelight contrasting with deep jade-cyan shadows, Kodak Vision3 500T)
     • Global mandatory visual suffix (style_suffix_en) : a standard English cinematography suffix of 25-40 words.
     • Global mandatory voiceover and audio lock (voice_and_audio_lock_en) : Since Google Flow supports direct inclusion of voiceover and sound effects, to prevent narration voice changes or music style shifts across video clips, you must define a fixed narrator voice persona (e.g., Consistent warm, deep, magnetic male Mandarin Chinese documentary narrator voice, calm poetic cadence, crisp studio condenser mic) and a unified background music direction.
   • Mandatory composition rule: style_suffix_en and voice_and_audio_lock_en must be directly merged and embedded into each shot's flow_video_prompt_en , ensuring the user can copy one prompt into Google Flow and generate coherent, stylistically unified, clean final footage with consistent voiceover.
   • ⚠️ No subtitle instruction: The prompt must never contain any instruction to generate subtitles or on-screen text (AI-generated video text is prone to distortion and ruins quality). Keep the Chinese voiceover script in a separate field for the user to generate high-quality subtitles later in editing software like CapCut or Premiere Pro.

Step 2: Total Duration Splitting & Generation Mode Decision (supports 15s one-shot or fast-cut storyboard)

Based on the user's specified total video duration (e.g., 30s split into 2 15s long takes, or multiple 5s-8s shots):

• 15s Compound One-Shot Mode (recommended for Google Flow 15s export) :
  • Split 30s into CLIP 01 (00:00 - 00:15) and CLIP 02 (00:15 - 00:30).
  • Inside each 15s prompt, plan a continuous, uninterrupted camera journey by sub-second intervals, e.g., [0-5s] -> [5-10s] -> [10-15s], and precise sound effect/voiceover timing points.
  • In the backup field (keyframe_image_prompt_en), provide an 8s Base + 7s SceneBuilder Extend (including segmented voiceover and sound effects) split prompt for flexible use in Google Flow.
• Intelligent generation mode badge (gen_mode_badge) :
  • T2V Text-to-Video (audio-visual integrated) (default 70%-85%): For landscapes, fluids, ink wash paintings, and atmospheric empty shots; consistency is maintained purely by the embedded global visual and audio locks.
  • I2V Image-to-Video (first frame/Ingredients) (15%-25%): Activated when the shot shows a specific product body, brand logo, or core cultural artifact; guides the user to upload a reference image as Ingredients.
  • F2V Frame-to-Frame bridging (key morphing transition): Used when two adjacent shots need a surreal, seamless morph transition.

Step 3: Seamless Transition Bridging Between Shots + All-in-One Audio-Visual Prompt Writing (clean visuals, no subtitles)

1. Visual Seamless Transition Bridge:
   • Select a transition from the 6 transition matrixes in references/cinematography_and_transitions.md (MATCH_SHAPE concentric circle match, WHIP_WIPE whip pan mask, SPEED_RAMP_DIVE macro penetration, LIGHT_BOKEH light bokeh breath, MOTION_RELAY motion relay, SCENE_EXTEND native extension).
   • The end of the previous shot must contain an Exit Momentum, and the start of the next shot must contain the corresponding Entry Momentum.
2. All-in-One Audio-Visual Prompt Structure:
   • For each shot, the flow_video_prompt_en must fuse the following modules into a single, complete prompt ready to copy into Google Flow (with no subtitle text instructions):
     1. [Visual & Continuous Camera] : sub-second shot descriptions and continuous camera movement;
     2. [Native Audio, Foley & BGM] : ambient sound effects (ASMR/Foley) strictly synchronized with the visual action, plus background music progression;
     3. [Voiceover (Mandarin Chinese)] : directly include the narrator voice description and the original Chinese voiceover script in double quotes (e.g., A warm, deep, magnetic male narrator speaks in standard Mandarin Chinese with poetic documentary cadence: "Yi long gaoling tu, luo bi cang shan he; qian du lie huo li, cui lian bu mie di wen run.");
     4. [Global Visual Lock] : the global camera and film stock style suffix lock.

Step 4: Dual-Mode Delivery (Markdown Script + Interactive HTML Director's Storyboard) & Cloud Sync

1. Organize the complete plan and storyboard data into a structured JSON, saved as /home/antigravity/flow_projects/<project_slug>.json.
2. Call the skill's built-in export script to generate local files and a visual HTML:

   python3 /shared/user/.gemini/skills/google-flow-director/scripts/export_storyboard.py \
     /home/antigravity/flow_projects/<project_slug>.json \
     --out-dir /home/antigravity/flow_projects \
     --html-artifact <ArtifactDirectoryPath>/<project_slug>_storyboard.html
   

3. Call the cloud sync script to automatically archive the generated .md storyboard script, .html interactive storyboard and .json source data to the user's Google Drive (Google_Flow_Director_Hub/01_Project Storyboards_Projects/<project_slug>/):

   python3 /shared/user/.gemini/skills/google-flow-director/scripts/sync_to_drive.py \
     --project-dir /home/antigravity/flow_projects
   

4. Also register <project_slug>_storyboard.html via write_to_file (with ArtifactMetadata: UserFacing: true, RequestFeedback: false) as a previewable Artifact, and summarize the core storyboard highlights and file/Drive links in the final response.