Video Enhance - Collect LTX Inputs

The Video Enhance node set is a two-track pipeline, and this is the node where the tracks meet. One track prepares the video and its guide anchors; the other enhances those anchors through Z-Image. VRGDG Video Enhance - Collect LTX Inputs waits for both to finish, validates that they actually did, fixes up conditioning positions that LTX would reject, and hands the LTX workflow every path and dimension it needs to do the real enhancement pass.

If you've ever wired a video-enhancement workflow and had it fail with a cryptic "missing file" error halfway through, you get why this node exists: it's a gate. It won't let you run LTX on a broken state, and it centralizes the inputs so you're not threading a dozen loose strings into the sampler.

The two inputs

Both inputs are the pack's VRGDG_VIDEO_ENHANCE_CONTEXT type - a structured bundle, not raw tensors:

• prepared_video_context - connect the Prepare Video and Anchors node. Its tooltip spells out the contract: connecting it proves the temporary full-frame LTX MP4 exists. If that file never got written, this node refuses to collect.
• enhanced_anchor_context - connect the Store Enhanced Anchors node. This proves all the Z-Image anchor results were saved. Both branches must be true before LTX is allowed to run.

The node also "adjusts LTX-incompatible conditioning positions" - meaning it normalizes anything in the prepared state that LTX would choke on, so the two branches can't disagree about coordinates or dimensions when they merge.

What it outputs

The LTX workflow's shopping list, all in one place:

• ltx_video_path - the temporary full-frame video LTX will enhance.
• enhanced_anchor_folder + anchor_indices + anchor_count - where the Z-Image-enhanced guide frames live, which frames they are, and how many. These feed LTX's anchor/conditioning inputs.
• ltx_width / ltx_height - the resolved working dimensions, exposed so downstream loaders match them exactly (no off-by-one-frame surprises).
• video_enhance_context - the combined context, passed on to the restore stage.

Where it sits

Chronologically it's the last node before the LTX pass: Prepare → Z-Image enhance anchors (via the Meta Batch loader and Store nodes) → Collect LTX Inputs → LTX generation → Restore Original Resolution. You can read it as the "render gate" - a clean place to pause, confirm both branches look right, and then commit to the expensive step. Because the whole point of this workflow is that LTX enhances using the Z-Image-enhanced frames as guides, getting the ordering right here is what prevents a half-prepared run from burning GPU hours on garbage.

Install

Standard pack install: ComfyUI Manager → search vrgamedev, restart, hard-refresh. This node needs the rest of the Video Enhance set and the VHS (Video Helper Suite) loader infrastructure that the pack's enhancement workflows depend on - if you're loading a shared workflow and it reports missing nodes, Manager will list them; the Video Enhance workflows bundle cleanly in the pack's Workflows/ folder.