VRGDG_UpdateLatestCombinedJsonPrompts_zimage

Some music video pipelines render stills first and animate them later. If yours does, this is the node you want - VRGDG_UpdateLatestCombinedJsonPrompts_zimage is the Z-Image-flavored version of the pack's prompt updater, and the _zimage suffix is the whole story. It edits the same combined-JSON prompt list, but it's built for the text-to-image pass that Z-Image (Alibaba's 6B image model) does to establish the scenes before they become video.

It exists because VRGameDevGirl's workflow has a Z-Image leg: generate the scene stills with a fast, high-quality image model, then animate them. Z-Image's strengths - speed, low VRAM, strong photorealism, and a LoRA ecosystem for keeping a character consistent across scenes - make it a natural fit for the "establish every scene as an image" stage of a music video. This node manages that stage's prompts.

How it works

Almost the same shape as the parent node, minus one toggle:

• combined_json_file - dropdown of the combined JSON files it can find.
• prompt_count (0 to 120) - how many prompt slots are active.
• remake_mode (default false) - force regeneration of the prompts instead of reusing the current file.
• folder_path (optional) - point at a specific folder if the dropdown's default misses.
• Up to 120 groups of prompt_number_N and prompt_text_N.

Note what's not here: the batch_type toggle (Text2Image/Image2Video) that the parent node has. This variant skips it, because the Z-Image path is the still-image pass - there's no mode to switch. You edit the text for each scene, the node writes the combined JSON, and the Z-Image side of the graph consumes it.

Outputs match the parent: updated_json_text, file_path, updated_count, and ignored (true when the node decided the file was already current).

Installing it

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

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

This node itself needs no models - but the Z-Image workflow it feeds absolutely does: a Z-Image checkpoint plus its Qwen3 text encoder. Those are on you to have already; the node just manages prompts.

Common issues

combined_json_file shows <no files found>. Same placeholder as the parent - no JSON exists in the expected spot yet. Generate one by running the workflow, or set folder_path.

ignored stays true and your edits vanish. Remake mode is off and the node considers the file current. Flip remake_mode to true to force a write.

You expected a batch_type switch. Look again - that's the parent node. This variant deliberately dropped it; it's the still-image leg and it knows it.

If your pipeline runs Z-Image scene stills before animating them, grab this over the generic updater. It's one less wrong-looking toggle on the node and one less way to accidentally switch the wrong pass.