Deng Studio · 登Studio · productions, 2026
Generated, avatar, programmed. Each film ships with its build record: the script, the measured timing, the model tests, the checks that had to pass. The workflow is the product.
Pick one to read it step by step
Working title: Old Brain, Young Money · documentary · dev cut 1 · 9:29
A nine-minute documentary made shot by shot from generated video. The narration is recorded first and becomes the clock for everything else; every quoted claim traces to an archived source.
Avatar presenter · 103-lesson course · one pipeline, run per video
An English sales course becomes Chinese lesson films. Transcript, glossary-bound translation, Abby's Fish Audio voice, a HeyGen avatar lip-synced to that exact audio, and a HyperFrames notebook page the lesson plays beside its English original.
Apple · Intention, bilingual edition (Coco) · Toastmasters Table Topic 「五」 (Cody)
The picture is code: one paused GSAP timeline inside HyperFrames, every element placed and timed from a measured file. One example pairs Chinese to Apple's typographic choreography frame by frame; the other builds a vertical recruiting clip from a generated image, a Fish voice and a lip-synced avatar.
How the studio works
The picture changes per workflow. The spine does not: a script with sources, a voice recorded before any picture, a HyperFrames edit generated from the voice timings, and a verification pass that reads the delivered file back.
Measure, never assume
A HeyGen clip's audio led its upload by 13 ms on one job and lagged by 23 ms on the next. A "quiet" sound effect came back near silent. Both were caught by an instrument, not by watching, and both became a step in the build script.
Sources or nothing
The documentary's quotes are audited against archived copies, all of them. The Apple edition takes its Chinese wording from Apple's own Chinese master and its motion from the English one, and writes down which file rules which.
One generator per film
Each film has a build script that reads the voice timeline and writes the composition. When a line is re-voiced, every graphic keyed to it moves with it. There is no second copy to drift.