RELAY — THE WARNING / Production cut 01
Original fiction by Vincenzo Romeo. AI-assisted production.

This is production evidence from the executed short, not a model-weight bundle.
No real actor, client commission, location shoot or commercial result is claimed.

CONTENTS
Ten selected motion-shot API graphs, exact input frames, actual decoded review
sheets, one native editor graph, fixed seeds, frame-addressed edit record,
English captions, the authored screenplay and the original score direction.
The home handset, two clock views and calling-HOME screen are four static phone inserts, not generated motion.
The cuff recognition take is cut before the later unmotivated expression drift.
The final cut uses one decisive physical knock, repeated from both perspectives.

RUN IN COMFYUI
Use ComfyUI 0.41.1 or a compatible newer version. Install the included small
custom_nodes/vcreative_shotkit folder in your own custom_nodes directory,
restart your own ComfyUI and copy input/* to its input directory. The editor
workflow uses native nodes plus VRReviewSheet. VRCampaignSpec is an additional
utility in that package, not a node used by this film's executed graph.

Load relay-workflow-ui.json in the editor, or execute a saved API graph with:
python3 render_one.py workflows/outside-call-api.json input/relay-outside-call.png
The runner talks to your local ComfyUI only; it never buys or downloads models.

MODEL DEPENDENCIES — NOT INCLUDED
ltx-2.5-22b-dev-transformer-comfy-int8-convrot.safetensors
ltx-2.5-22b-distilled-transformer-comfy-int8-convrot.safetensors
gemma4-12b-with-proj-ltx-2.5-comfy-int8-convrot.safetensors
ltx-2.5-video-vae-bf16.safetensors
ltx-2.5-audio-vae-bf16.safetensors
ltx-2.5-latent-spatial-upscaler-x2-bf16-1.0.safetensors
Review each provider's current model licence and hardware requirements.
Executed locally with 24 GB VRAM. Identical seeds across different hardware or
software versions do not guarantee bit-identical results.

AUDIO
The warning is native generated LTX dialogue, reused exactly: telephone-filtered
on reception and full-band when the source is revealed. The final plea is a
separate native performance. Spoken words were checked with local ASR; this
does not certify subjective voice quality. Foley comes from generated takes
and an authored room-tone layer. The score is an original Suno V6 instrumental
through the owner's existing SunoAPI.org integration. Private provider receipts
and unedited music masters are retained outside this public kit. No new music
request or account credentials are included in the runnable ComfyUI graph.

OUTPUT
1536 x 864 / 24 fps / H.264 + AAC. Learned 2x latent refinement precedes the edit.
Reference images were generated/edited with OpenAI Images. No beauty filter,
sharpening pass or grain overlay was used to conceal a weak performance.
All material remains a local review cut until approved for publication.
