Production Guide

Workflow Overview

Give the system a storyboard grid and product photos, and it produces 9 ready-to-edit video clips — one per storyboard cell. Uses Gemini to interpret the storyboard and write motion prompts, then sends those prompts to Kling V3.

Stage 1 — Storyboard Analysis (Gemini)

A single 3x3 grid storyboard image is loaded and fed into a Gemini 3 Pro Preview node. The system prompt instructs Gemini to act as a motion director for high-end product ads.

What Gemini does:

• Reads the 9-cell storyboard (left-to-right, top-to-bottom → Shot 01–09)
• Writes a Kling-compatible image-to-video motion prompt for each shot
• Outputs structured XML containing a <shotN_prompt> and <shotN_duration> for each shot

System prompt rules:

• Describe ONLY camera movement and light changes (Kling already sees the image)
• Very slow, subtle movements only — product must not deform
• Close-ups → slow orbit + subtle light shift
• Medium shots → slow arc (~15°) or slow pull out
• Wide shots → slow pull out or slow dolly forward
• Never use: fast, zoom, spin, rotate 360, handheld, shaky
• Each prompt is 1–2 sentences max
• Movements must be varied across shots

Example: 3×3 storyboard grid — each cell maps to one shot

Output format (XML):

<shot1_duration>5</shot1_duration>
<shot1_prompt>Very slow orbit around product. Subtle lighting shift...</shot1_prompt>
<shot2_duration>5</shot2_duration>
<shot2_prompt>...</shot2_prompt>
...through shot 9

Stage 2 — Prompt Parsing

The Gemini output is a single block of structured XML text. A custom parser node splits it into 9 individual prompt strings. Each output is routed to its corresponding Kling generation node.

Note: The Gemini output is previewed via a "Show Anything" node. The structured text is currently entered into a DF_Text node which feeds the parser. There may be a manual copy-paste step between Gemini output and DF_Text input.

Each parsed output is also routed to a PreviewAny display node (labelled "Shot N: # of Seconds") so you can verify per-shot prompt and duration before generation runs.

Stage 3 — Video Generation (Kling V3 x 9)

Nine parallel KlingVideoNode instances each receive a start frame image and a motion prompt.

Shot → Start Frame Mapping:

Example: 9 generated video frames from Kling V3

Stage 4 — Save

Each of the 9 Kling outputs is connected to a dedicated SaveVideo node, saving clips to video/ComfyUI/ with automatic naming.

Data Flow Diagram

┌─────────────────┐      ┌──────────────┐      ┌────────────────┐
│  Storyboard      │─────▶│  Gemini 3    │─────▶│  Show Anything │
│  (3×3 grid)      │      │  Pro Preview │      └────────────────┘
└─────────────────┘      └──────┬───────┘
                                │
                                ▼
                         ┌──────────────┐
                         │  DF_Text      │
                         └──────┬───────┘
                                │
                                ▼
                         ┌──────────────┐
                         │  XML Parser   │ (splits into 9 outputs)
                         └──┬──┬──┬──┬──┘
                            │  │  │  │  ... (×9)
                            ▼  ▼  ▼  ▼
┌──────────────┐     ┌─────────────────┐     ┌─────────────┐
│ Product Image │────▶│  Kling V3 Node  │────▶│  SaveVideo   │
│ (per shot)    │     │  (per shot)     │     │  (per shot)  │
└──────────────┘     └─────────────────┘     └─────────────┘

How to Use

1. Prepare your storyboard — Create a 3×3 grid image where each cell represents one shot.
2. Load product images — Place 9 product photographs into the corresponding LoadImage nodes.
3. Run the Gemini stage — The storyboard is analysed and motion prompts are generated. Review the output.
4. Transfer prompts to DF_Text — Copy the Gemini XML output into the DF_Text node (or verify it's wired directly).
5. Run Kling generation — All 9 video clips generate in parallel. Output saves to video/ComfyUI/.
6. Post-production — Import the 9 clips into your NLE (Premiere, DaVinci, etc.) and assemble the final sequence.

Key Customisation Points

• System prompt (Node 5) — Edit to change motion style, add/remove movement types, adjust pacing rules
• Product images — Swap out per-shot start frames for different products or angles
• Kling settings — Adjust resolution, aspect ratio, or duration on the KlingVideoNodes
• Storyboard — Change the input grid to get completely different motion direction from Gemini

Final Output

The 9 clips assembled into a finished product video: