Studio
A small studio
with a long shot list.
Zhulu is a two-desk production studio working in generated film: short-form drama, animated serials and brand films, made end to end — script, boards, generation, edit, grade, mix.
We are not a prompt marketplace and not an agency with a render farm. The studio exists because generated footage is only useful when it behaves like footage: the same face in shot nineteen as in shot one, a product that does not change shape between cuts, a machine that an engineer recognises.
That is a production problem, not a model problem, and it is solved with boards, references, take counts and a lot of rejected frames.
Pipeline
Development. Script adaptation, beat sheet, runtime and shot count. The edit is planned here; generation only fills it.
Boards and identity. Character sheets with three angles per face, location frames, wardrobe and product references. Nothing is generated until these are approved.
Generation. 20–25 steps at production resolution on self-hosted ComfyUI pipelines. Each shot carries its own reference slots and one primary movement.
QC. Frame-level review of hands, faces, signage and product; automated transcription checks against the script; loudness and true-peak measurement at the master.
Finish. Grade, mix, subtitles, AI disclosure, upscale where the platform needs 4K, and a clean master for re-versioning.
Stack
Self-hosted ComfyUI pipelines on our own GPUs, with MiniMax H3 for native dialogue and lip-sync, Wan 2.2 and LTX-2 for coverage and style work, and SeedVR2 for upscale. An in-house studio tool handles job queues, takes and delivery records.
Every project keeps a per-shot production record: model version, prompt text, seed, canvas, steps and the reason for each re-take. Series work continues from those records rather than from memory.
Standards
No single-frame sign-off. A shot is judged across its frames, not on a still. Dialogue is checked by transcription, not by ear alone.
Masters, not exports. H.264 high profile, Rec.709, −16 LUFS integrated, true peak ≤ −1 dBTP, audio at 48 kHz.
Reproducibility. Any delivered shot can be regenerated from its record by another operator.
Disclosure
All footage on this site is AI-generated and carries the studio's own AI badge or platform disclosure where required. Nothing here is presented as live-action photography.
We do not use third-party footage as a generation base, do not reproduce a real person's likeness or voice without written permission, and do not remove disclosure on delivery.
Availability
China, UTC+8, working with clients in North America and Europe. Enquiries are answered within 12 hours; calls are scheduled to overlap with your morning.
English and 中文 for production; English for delivery documentation.