Kling 3.0 1 Click Multishot (hellorob): a Gemini shot planner, regex tag splitting and six Kling storyboards

What this is: a reading of one community workflow file in the Comfy-Org workflow_templates repository at commit 0cc0fba5, Kling 3.0: 1 Click Multishot, credited to the community creator hellorob (file templates_rob_kling3_0_multishot_llm_product.json, saved by ComfyUI frontend 1.54.15). Only the JSON was inspected. No workflow was run, no Gemini or Kling call was made, and the example images and output video were not opened. Declared pipeline: 1. Controls. "Number of Seconds (3-15)" stores 8 and "Number of Shots (1-6)" stores 2. Seconds feeds both the Kling node's duration input and the prompt builder; shots feeds only the prompt builder. 2. Prompt builder (subgraph Create Prompt). A Text Multiline template reads: "Use the input image 1, create a prompt that has {N_SHOTS} distinct shots. The total time for the shots is {N_SECONDS}." followed by {USER_STORY}. Two RegexReplace nodes (case-insensitive, all matches) replace the pattern N_SHOTS with the shot count and N_SECONDS with the seconds, each converted to text by a KJNodes SomethingToString node. 3. Shot planner. GeminiNodeV3 (Gemini 3.8 Flash, thinking MEDIUM, up to 32768 output tokens, seed 42 with randomize) gets that text as its prompt and only the start-frame image as image 1. Its stored system prompt (4,150 characters) casts it as a creative director. It must read the product image and lock the background as a constant for every shot. It maps the shot count to an ad arc (Hero, Detail, Feature, Lifestyle, Outro; for example 3 shots are Hero, Feature, Outro). It assigns whole-second durations that sum exactly to the total (at most 15 s), by default about 25% to the Hero shot and 20% to the Outro. Shot 1 is image-to-video and describes only motion from the start frame; later shots are self-contained text-to-video prompts in the same environment, 40-80 words each, with no marketing copy. The output must be only tagged blocks: shotN_duration and shotN_prompt tags for each shot. 4. Manual override (subgraph Selected Prompt). A StringCompare checks whether the "Prompt Input" box equals an empty string; a switch then passes Gemini's text when the box is empty and the pasted text otherwise. "Preview All Prompts" shows Gemini's raw output. 5. Splitting (subgraph Extract Prompts). Twelve RegexExtract nodes hold one prompt pattern, (.+) between shotN_prompt tags, and one duration pattern, digits between shotN_duration tags, for each shot 1-6. All use First Group, group 1, case-insensitive, multiline on and dotall off. Durations pass through StringToInt (string-converter pack). Prompts and durations go to the Preview Prompts nodes and, through reroutes, to the storyboard 1-6 prompt and duration inputs of the Kling node. 6. Video (group Generate and Save Video, bypassed in the saved file). KlingOmniProImageToVideoNode uses model kling-v3-omni at 9:16 and 1080p, storyboards set to 6, generate_audio true and seed 1264338454 with randomize. Its own fixed prompt begins "Use image 1 as a start frame and the other images as product reference for fine details." Its reference images are the start frame plus the two product images, each product image resized and padded with white to the start frame's size and then batched (subgraph Resize and Batch Images). SaveVideo writes with prefix video/ComfyUI, format and codec auto. The workflow's own note gives the intended use: run once to generate prompts; copy from Preview All Prompts into Prompt Input and bypass the LLM group; check the per-shot previews; unbypass Generate and Save Video, set storyboards to the shot count; run again. Reading notes (inferred from stored settings, not run): - The regex replaces only the letters N_SHOTS and N_SECONDS, so the braces stay, and the text sent to Gemini would read "has {2} distinct shots" and "is {8}". No node fills {USER_STORY}, so it reaches Gemini literally unless the template text is edited. The system prompt says to rely on the image when no story is given. - Gemini sees only the start frame. The two product images reach Kling only, as references. - With dotall off, (.+) cannot cross a line break, so a shot prompt that Gemini splits over lines would not be extracted. - As saved, storyboards is 6 while shots is 2; the note tells users to make them match. What happens to empty storyboards 3-6 is not established. - The Kling duration widget stores 3, but its input is linked to the seconds counter, and the stored storyboard durations of 4 are likewise replaced by the extracted values. Node packs named in the JSON: SomethingToString from comfyui-kjnodes, Text Multiline from was-ns, StringToInt from comfyui-string-converter; the other nodes with a pack id are comfy-core 0.14.1. Not established: what Gemini actually writes, whether its durations add up, how the Kling service treats the storyboards, what the output video or audio looks or sounds like, and the cost of a run. The example images and output video listed for this template were not opened. Kling and Gemini service terms are not covered.

{
  "citations": [
    {
      "url": "https://raw.githubusercontent.com/Comfy-Org/workflow_templates/0cc0fba562efe9f68e48050c2b82ed6495050bc2/templates/templates_rob_kling3_0_multishot_llm_product.json",
      "note": "Pinned original; exact hashes and byte ranges retained in source pins."
    },
    {
      "url": "https://raw.githubusercontent.com/Comfy-Org/workflow_templates/0cc0fba562efe9f68e48050c2b82ed6495050bc2/LICENSE",
      "note": "Pinned original; exact hashes and byte ranges retained in source pins."
    }
  ]
}