VRGDG_UpdateLatestCombinedJsonPrompts

At the heart of VRGameDevGirl's music video workflow there's a JSON file holding every scene's prompt - number, text, and which image each one anchors to. Editing that file by hand is a misery: one misplaced comma and the whole batch dies. VRGDG_UpdateLatestCombinedJsonPrompts is the node that edits it for you, so you can tweak scene 12's prompt from inside the graph and regenerate just that chunk.

It's the control panel for the pack's batch generation. The workflow turns your lyrics and concepts into a list of scene prompts, writes them into a "combined JSON" file, and then this node is how you come back and adjust that list - fix a character's outfit, swap which reference image a scene uses, or flip the whole batch between a text-to-image pass and an image-to-video pass.

How it works

The essential inputs:

• combined_json_file - a dropdown that lists the combined JSON files it can find. Select the one you're editing.
• prompt_count (0 to 120) - how many prompt slots are active. Set this to your scene count and the node works through them.
• batch_type - Text2Image or Image2Video. This is the big toggle: the same prompt list can drive a still-image pass or a video pass, and this node rewrites the JSON for whichever one you're running.
• remake_mode (default false) - flip it on to force the prompts to be regenerated rather than reused.
• folder_path (optional) - point at a specific folder if you're not using the dropdown's default location.

Then there are up to 120 groups of three fields each - prompt_number_N, prompt_text_N, prompt_image_index_N - one group per scene. The text is the actual prompt, and the image index says which generated still (or reference image) that scene is anchored to. The node writes your edits back into the JSON.

It returns four outputs: updated_json_text (the whole file as a string, handy for wiring downstream), file_path (where it was written), updated_count (how many prompts actually changed), and ignored (a boolean - when the node decides nothing needs to change, e.g. remake mode is off and the file is already current).

Installing it

Ships in the VRGameDevGirl Video Enhancement Nodes pack. ComfyUI Manager → search vrgamedev → install, restart ComfyUI, or:

cd ComfyUI/custom_nodes
git clone https://github.com/vrgamegirl19/comfyui-vrgamedevgirl

No models to download. If the pack's requirements.txt (kornia, librosa, imageio) wasn't installed by Manager, pip install -r requirements.txt against your ComfyUI Python.

Common issues

combined_json_file shows <no files found>. That's the node's own placeholder default - it means no combined JSON exists in the expected location yet. Run the workflow once to generate one, or point folder_path at where your file actually lives.

The ignored output is true and nothing updated. The node is refusing to touch a file it thinks is already current. If you want changes written, flip remake_mode to true.

You set prompt_count and the extra slots do nothing. Only slots 1..prompt_count are live; the rest are ignored by design. If you need scene 200, raise the count toward the max of 120.

It's a lot of surface area for one node - 120 numbered slot groups is intimidating on first open. Ignore the wall of fields; you'll use maybe six slots and prompt_count. Once it clicks, it's the difference between editing one JSON by hand and adjusting your whole video from the graph.