Zura · Load Reviewed Shot

Most "load a clip" nodes will happily load whatever path you give them. GenjLoadReviewedShot ("Zura · Load Reviewed Shot") is stricter in a way that's worth understanding, because it's the node that stands between the mask you approved and the 20-minute render you're about to start.

Two IDs, and why they're separate

It takes shot_id (the reviewed shot from GenjSaveReview) and approved_shot_id. The second one exists because Studio fills in the approval after you click through the review, and the node refuses to run unless the two agree:

Approve the reviewed mask in Zura Studio before generation. Standalone graphs require matching reviewed and approved shot IDs.

In a hand-built graph, that means wiring the same string into both fields. It's a two-field handshake you'd normally do with a boolean, but a boolean can't tell you which shot you approved - and that's the failure this is designed to prevent.

Before returning anything it re-reads review.json and re-hashes every file in the shot folder, then re-hashes the original source video against the hash recorded at selection time. Change the clip on disk, re-encode it, or move it, and you get The original source changed. Prepare and review the shot again. Same story for a reference audio track: The reference audio changed. Prepare the shot again. The node also sets IS_CHANGED to a NaN sentinel when the IDs don't match, which forces ComfyUI to re-execute it rather than serving a cached result.

The three optional knobs

• render_long_edge (0–1920, default 0). 0 uses the reviewed preview. Set it to 1280 or 1920 and the node re-reads the original footage at full resolution, scales it with lanczos, and resizes the approved mask to match - so you review at 768 and render at 1280 without losing the approval. It's cached per size, so the re-read only happens once. Note the model canvas is rounded to a 64-pixel grid here, unlike the even-pixel rounding in GenjSelectClip, because this raster feeds a diffusion model.
• output_scale (1 or 2). 2 halves the model canvas, which is the escape hatch when you're about to blow out VRAM. Delivery resolution is tracked separately and restored at export.
• preview_seconds (0–30). Trims everything - source, mask, guide - down to a short test take from the start of the shot. This is what the pack's H3 two-second preview uses.

Outputs

Five of them, and the names are honest:

• source_clip - a VIDEO of the reviewed footage at your chosen render size.
• performer_mask - a video (not a MASK), the approved combined mask rendered as frames. It's what you feed to the nodes that expect a mask video.
• seedance_guide - the guide.mp4 from the review: guidance frames plus the guidance audio, ready as a reference clip for a hosted model.
• clip_details - GENJ_CLIP, now carrying review_shot_id and the export raster. Everything downstream that needs to prove provenance reads this.
• genuine_performer_mask - the separately saved performance-only mask, or the combined mask as a fallback if the review didn't include one.

Install

This node ships with Zura Nodes: ComfyUI Manager → search Zura Nodes → install → restart. Manual install:

cd ComfyUI/custom_nodes
git clone https://github.com/ZURAVFX/ComfyUI_zura_nodes
python -m pip install -r ComfyUI/custom_nodes/ComfyUI_zura_nodes/requirements.txt

PyAV and ffmpeg on PATH do the re-read and rescale; there's no model involved, so this one works on a box that has the pack installed and nothing else downloaded.

Gotchas

Don't delete the shot folder. The approved mask lives in output/genj/reviews/<shot_id>/, and clearing your outputs breaks the gate with an unhelpful-looking file error. If you're short on disk, delete takes from genj/final, not genj/reviews.

Mismatched IDs after an edit. If you tweak the mask and re-run GenjSaveReview, the shot ID changes and the old approval no longer validates. That's the feature working; paste the new ID.

Expecting a MASK from performer_mask. It's a video. If a node wants a MASK, that's what ZuraWanReviewedFrames and friends are for - a very common wiring mistake when you're reading the outputs cold.