HUDHUD Technology · DOXA System
DOXA is Hudhud's in-house production agent. It coordinates planning, references, generation and revisions inside the tools the team already uses. People approve every paid or irreversible action.
DOXA lives inside Slack. The team asks in plain language and drops in a storyboard, a script or a reference. DOXA reads it, plans the work, runs the research and generation, checks the result and files it by type. Anything that costs credits waits for a person to press Approve.
The agent coordinates the workflow. The proprietary technology is the agent and workflow layer, not a foundation model built from scratch. DOXA is designed to use the best available generation tools, and to keep the director in control of the result.
Human approval is the feature. Every paid or irreversible action waits for a person.
“Do not fabricate history with AI.”
DOXA, production note
That note is from a real production. A foundation campaign needed 1945 archive material. DOXA separated the real records from the generated dream sequences and refused to invent historical documents. The pipeline can generate a world. It should not generate a lie.
DOXA is a company-described production agent. The workflow is documented in the U.N.D.A.Y. creative development log. It is not an audited benchmark or a proprietary foundation model.
Automation with approval control.
A brief and source files arrive in Slack, in plain language.
DOXA plans the work, runs the research and reports progress in the channel.
People authorize generation credits and irreversible actions. A plan card shows the references, settings, prompts and estimated cost before anything is submitted.
DOXA checks the results and files them by asset type: documents, images, video, references.
There are two ways to make a film with DOXA. The right one depends on what the story needs.
A main script goes in. A complete film comes out. The team supplies the script and direction; DOXA generates every scene. RECEIVER is the proof: 7:42, a festival screening in Chile, a reported selection in Shanghai.
A real performance stays at the centre. AI builds the world around it: locations, backgrounds, vehicles and action. The actor is filmed; the director controls the shot and the final frame. Pink Truck is the demonstration.
Full Auto is the purest form of the pipeline. Hybrid is the version most studios need: the performance is real, the world is generated, and the actor is still the reason the scene works.
Full Auto is demonstrated on RECEIVER. Hybrid is demonstrated in the performance-transfer comparison and the Pink Truck shot design. Both are company-supplied demonstrations, not a guarantee of identical results on every script.
Every stage produces material the director can review and revise. Change the script, and the downstream work can be rebuilt.
Translate the story into shots, lenses, lighting and camera movement.
Define characters, locations, wardrobe and props.
Build shot sequences from the script and the approved designs.
Generate moving images from the approved boards.
Integrate filmed performances with the generated world.
One working session, captured on one screen: the board, the shot list, production design, 3D blocking and the video stage. The recording runs 56 seconds; sound and playback controls sit under the screen.
DOXA / Console
Screen capture 00:56
00:00 00:56
The pipeline is only interesting if it has made something. These are the projects that test it.
A 7:42 complete AI film, made from a main script. Online screening at IA en Corto, Chile, October 2026. Selected for screening at the Shanghai AI & Robot Film Festival, as reported by the production team.
A foundation campaign developed through a 76-second animatic, then a 30-second teaser package with 24 cuts, 122 look references, 49 creative techniques and a working approval flow. The final 30-second teaser remains in development.
An AI Hybrid feature based on Jang Hang-jun’s original screenplay. About 80 minutes, eight-month schedule. The reference model for the studio’s feature pipeline.
Delivered commercial work: USP cuts, a vehicle teaser and a highlight film. Filmed vehicle geometry anchors generated branding, headlamp and interior variants.
Episodic drama produced through the Full AI pipeline, with no physical set or shooting crew. Consistent characters and production design across episodes.
RECEIVER's Chile screening is public. The Shanghai selection is reported by the production team. Pink Truck runtime and schedule are planning estimates. ZEEKR and DramaBox are company-reported commercial work.
The model replaces the parts of production that scale with headcount and locations: sets, crowds, repeated setups, reshooting a detail. It keeps the parts that scale with taste: the script, the direction, the performance, the final frame.
Request the production modelThe detailed cost model is shared in partner conversations. It is a planning model, not an audited saving.
For studios, broadcasters and production companies that want to build an AI-native pipeline of their own. The programme is practical: the team learns on real scenes, compares the output with the original footage, and leaves with the workflow, the checklist and the cost model.
Select and test the bridge, transformation and action samples.
Planning, references, storyboards, filming, generation, compositing, editing and finishing.
3D camera design, background and wardrobe transformation, AI action and face replacement.
Compare the 3D design and the original footage with the AI result.
Compare like-for-like quality and duration, including rework and post.
Pipeline map, training examples, comparison table and quality checklist.
The programme is a training service. Schedules, fees and deliverables are agreed per engagement.
Tell us what you want to make. We will show you the route, the cost and the approval points.
