ShengShu Technology (Vidu) is hiring: Data Strategy Expert (Screenwriting & Directing) (数据策略专家 (编剧导演)) (Beijing, China). ShengShu Technology is the Beijing-based company behind Vidu, an AI video-generation model and creation platform. Main focus: Define the video caption system: For the training needs of the video generation model, formulate the standard of "how a video should be described" - clarify which dimensions the caption needs to cover (action and motivation, plot advancement, scene and scheduling, camera movement, timing relationship, subject and environment, etc.), as well as what granularity and language style should be used for each dimension, and the output will be implementable caption writing specifications and annotation SOP, which can be directly passed to the annotation team for execution.
Key responsibilities
- Define the video caption system: For the training needs of the video generation model, formulate the standard of "how a video should be described" - clarify which dimensions the caption needs to cover (action and motivation, plot advancement, scene and scheduling, camera movement, timing relationship, subject and environment, etc.), as well as what granularity and language style should be used for each dimension, and the output will be implementable caption writing specifications and annotation SOP, which can be directly passed to the annotation team for execution;
- Establish Caption quality inspection standards: define the criteria for determining whether a caption is qualified or not (such as whether the description is accurate, whether key actions/plots are missing, whether the actions correspond to the intention, whether the timing is correct, whether the language conforms to the specifications, whether there are hallucinations/excessive brain-filling, etc.); establish a random inspection mechanism to continuously locate shortcomings in annotation quality and feed back the iteration of caption specifications;
- Promote the quantitative implementation of standards: Collaborate with the quantification and engineering teams to translate caption writing specifications and quality inspection standards into statistical and automated quality inspection rules and indicators (such as coverage, accuracy, prompt-video semantic alignment, timing/action consistency, etc.) to improve quality inspection efficiency;
- Cross-team alignment: Collaborate with algorithm, product, and evaluation teams to understand the model’s true demands for captions (such as instructions to follow, complex action sequences, description methods required for multi-shot narratives), promote the optimization of the caption system, and evaluate its actual impact on model performance.
Requirements / qualifications
- Professional-level narrative and picture interpretation skills: In-depth understanding of script structure, character motivations, dramatic conflicts, scene scheduling and lens language, and the ability to accurately understand what is being performed and how it is performed in a video from the professional perspective of a screenwriter/director, and explain it clearly. Those with professional backgrounds in screenwriting/directing from top film and television schools will be given priority, but actual description and dismantling abilities will be more important than award-winning or work resumes.
- Relevant experience: Experience in screenwriting, directing, script planning, storyboarding/storyboarding, film and television content planning, etc.; experience in formulating annotation specifications, content description/marking, and quality control caliber design is particularly valued; candidates with experience in formulating AI video annotation/caption SOPs will be given priority;
- Be able to write description standards into rules: "How should a video be described?" can be broken down into formulaic standards that are interpretable, reusable, and can be passed to the annotation team for execution - so that the captions written by different people for the same video are as consistent as possible. This is the core competency of this position;
Nice to have
- Bonus points: a. Proficient in using vibe coding / AI tools to improve work efficiency, and actively explore and learn new skills; b. Familiar with mainstream video generation models such as Seedance, Kling, Happyhorse, etc.), and have an independent understanding of "what kind of captions are needed for the model to learn well".
Team: R&D Location: Beijing, China Employment type: Full-time Posted date: 2026-08-14 Original title: 数据策略专家 (编剧导演) Original posting is in Chinese; description translated to English.
