Boyo Storyboard Output

This is the primary decoder in the Boyonodes storyboard chain: paste the JSON an LLM produced for your storyboard, and it splits into 12 individual prompt outputs - image_scene1 through image_scene6 and video_scene1 through video_scene6 - ready to wire into image and video generation nodes. Where it goes beyond a dumb JSON splitter is the six optional speech_scene1–6 inputs, which let it re-embed dialogue back into the video prompts for the pack's OVI audio workflow.

How it works

Feed it the raw JSON string in json_input, and it parses the scene structure, mapping each scene's image prompt and video prompt to the matching output. The speech inputs are the interesting part: in OVI (audio) mode, the video model needs the spoken dialogue woven into the prompt - typically wrapped in tags the model recognises. If you provide speech_sceneN alongside the JSON, the node integrates that speech back into the corresponding video_sceneN output so the video generator knows what's being said during that scene. Leave the speech inputs empty and it behaves like a straight 12-way splitter.

The outputs are plain strings, so from here each scene prompt can go to its own sampler, a switch node, or a batch collector - the storyboard becomes a normal graph of prompts instead of one giant blob.

Where it fits

The full pack chain is: BoyoStoryboardPrompt builds the instruction prompt for an ollama model (which supports a "System Prompt 3 (OVI Audio)" mode) → the LLM returns JSON → BoyoStoryboardOutput splits it into per-scene image and video prompts → image prompts feed your image editor (Qwen-Image edit / HiDream / Kontext-class models) and video prompts feed your video model (Wan). With speech filled in, the video prompts carry the dialogue so a subsequent TTS/audio step lines up.

It shares the 12-output structure with BoyoStoryboardJsonParser - the alternative parser for when one of them chokes on a particular model's JSON formatting.

Installing it

Ships with Boyonodes (DragonDiffusionbyBoyo):

cd ComfyUI/custom_nodes
git clone https://github.com/DragonDiffusionbyBoyo/Boyonodes

Restart ComfyUI or install "Boyonodes" via ComfyUI Manager. No extra dependencies.

Gotchas

• The JSON must be the model's raw output, not wrapped in markdown code fences - strip those before pasting or the parser rejects it.
• Speech inputs are only meaningful if the system prompt you used generated an OVI-audio storyboard. Feeding speech into a standard 6-scene JSON just appends dialogue text where the video model may not expect it.
• It assumes six scenes; anything else leaves outputs empty or truncated.

If you're building LLM-planned multi-scene videos, this is the seam where unstructured AI output becomes a workable ComfyUI graph - and the speech inputs are the part most generic parsers won't give you.