Face Fix - Collect LTX Inputs

The Face Fix pipeline is a relay race, and VRGDGFaceFixLTXInputs is the checkpoint where the baton actually has to be in hand. The author calls it "an execution barrier" - it validates that two expensive prerequisites really happened, then exposes the results as clean inputs LTX can consume. If the cropped face video wasn't created, or the enhanced anchors weren't saved, this node stops the race instead of letting LTX churn on garbage.

How it works. Two inputs, both face_fix_context connections, each proving a distinct milestone:

• cropped_video_context - from Create Cropped Face Video. This is the proof the 512×512 source MP4 was actually created. The tooltip is explicit that the proof is the point.
• enhanced_anchor_context - from Store Enhanced Anchors. The proof that all required Z-Image anchors were validated and saved.

Because these contexts come from the tail of their respective sub-pipelines, wiring them in forces ComfyUI to run those pipelines to completion before this node (and everything downstream of it) executes. That's the "barrier" part - it converts "I hope the anchors were done" into "I will not proceed until they are." It's the same ordering trick the pack uses everywhere (the any/* triggers on the general nodes), but here it's doing real validation work, not just sequencing.

The outputs. Once both prerequisites pass, the node hands LTX everything it needs:

• cropped_face_video_path - where the 512×512 face MP4 lives, for LTX's video-to-video conditioning.
• enhanced_anchor_folder - where the enhanced anchors are, for identity conditioning.
• anchor_indices - which source frames the anchors correspond to.
• anchor_count - how many, so LTX config knows what it's working with.
• face_fix_context - passed through to the composite stage at the end.

Why you'd reach for it. When a pipeline has multiple stages and one of them costs you real GPU hours (Z-Image on every anchor plus LTX regeneration), you want the stages strictly gated. This node is the difference between "the graph is correct in theory" and "the graph cannot run out of order." It's also the node that makes the workflow importable by beginners - they can't accidentally wire LTX before the prep work exists, because the inputs won't accept the wrong stage's context.

Installing. In the vrgamedev pack:

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

or install via ComfyUI Manager, with VideoHelperSuite and the LTX 2.3 model set alongside.

Troubleshooting. If this node errors out, it's almost never broken - it's doing its job. A missing context means the upstream stage didn't produce what it promised: check Create Cropped Face Video if cropped_video_context won't connect, or Store Enhanced Anchors if the anchor side fails. And don't try to feed it a context from the wrong stage just to make the wire connect - the validation exists because running LTX without those artifacts produces a video you'll throw away.