JSON Shot Splitter

Here's a workflow that comes up a lot once you have an LLM planning your videos: the model returns one big JSON blob describing every shot in a sequence, and you need to turn that into individual prompts, one per sampler node. That's exactly what JSON Shot Splitter does - and it's the unsung workhorse of the pack's N8N story.

It takes a JSON string and a shot_count, reads a shot_manifest object out of it (a dict with shot_1, shot_2, ... keys), and splits each entry into its own output socket: shot_1 through shot_20. If an entry is already a string it passes through untouched; if it's an object or array it's serialized to pretty-printed JSON so you can still read it.

The inputs that matter

• json_string - the manifest JSON. It's a forced input (forceInput), meaning you'll typically wire it from an LLM node or an N8N response rather than typing it, which is the right call - pasting a big manifest by hand is miserable.
• shot_count - 1 to 20, default 2. This controls how many manifest entries get split out. A neat detail: the pack's JS extension syncs the visible output sockets to this value, so at shot_count: 3 you see three sockets, not twenty. The extra outputs still exist in the schema and just come back empty.

Outputs: shot_1 … shot_20 (all STRING).

Installing

The usual:

cd ComfyUI/custom_nodes
git clone https://github.com/studiodeep-ai/comfyui-studiodeep StudioDeep
pip install -r StudioDeep/requirements.txt

or search "StudioDeep" in ComfyUI Manager, then restart. No keys, no models.

Troubleshooting

• "Invalid JSON string" - the input isn't valid JSON. If it's coming from an LLM, watch for markdown fences or trailing prose; those break json.loads.
• Empty outputs with valid JSON - your JSON has no shot_manifest key (the node looks specifically for shot_manifest.shot_N), or the keys are named differently. Rename the keys in whatever produces the JSON and the node will find them.
• A shot comes back as JSON instead of text - that's by design for object entries. If you wanted the raw string, the upstream JSON needs the shot as a string in the first place.

The honest take: it's a small, opinionated node with a specific contract (shot_manifest), and it's clearly built to sit downstream of an n8n workflow that produces that shape. If you control the workflow, this makes the LLM-to-multiple-samplers hop a one-wire affair; if your JSON has a different structure, you'll be reformatting it first.