LOCAL MCP / ACTUAL GENERATION + NATIVE EXPORT

A brief. A reference.
A take you can edit.

An agent created this unbranded headphone reference, inspected it, rendered a gentle camera push, then paired the result with a measured local voice and editable word highlights. The image and video were generated through Aftertake’s native Comfy SDK workflows.

01 / Actual LTX take · 768 × 768 · 97 frames / 24 fps · Original Z-Image reference ↗
02 / Native Film export · 720 × 720 · 2.5 seconds · complete local voice + editable type
01 / CREATE AND LOOK

The actual reference.

Z-Image Turbo generated the product still locally. The MCP client decoded and inspected it before submitting the video recipe.

02 / KEEP THE SAME JOB

Reconnect without duplicating.

The same image and take requests returned the original native IDs after a studio restart. Their saved graphs and seeds stay attached to the output.

03 / FINISH IN FILM

Words with their voice.

One complete Kokoro line, five retained word timings and editable type. The native export contains the matching project and verified source hashes.

Download narrated proof ↧Editable project ↧Source + original media ↗MCP tool guide ↗

The same cut, back in the editor.

This is the actual reopened Film project. The take, complete voice and speech-following typography remain editable. The source packages include a companion project that imports the exact take bytes into a fresh CPU editing workspace.

Actual Aftertake Film editor showing the generated headphone shot, editable type and one-cut timeline

Recorded local proof on an RTX 5080. The public page plays finished media; new generation needs the local studio, Comfy engine and model weights. One short product shot does not establish general identity consistency, long-form quality or full Higgsfield parity. Measured evidence. Fictional unbranded concept, generic local narration, no music or identity cloning. Model licenses apply separately; app code is MIT. Final review and submission remain with Jay.