Skip to content

Projects and Works

AutoVio organizes everything in a two-level hierarchy: projects and works.

A project is the top-level container. It holds:

  • Name and metadata
  • Project type — Determines default prompts and style: blank, saas, news, social, ecommerce, educational
  • Style guide — Tone, color palette, tempo, camera style, brand voice, must-include / must-avoid (used in scenario, image, and video generation)
  • System prompts — Default scenario prompt, analyzer prompt, image instruction, video instruction (all can be overridden per work)
  • Knowledge — Extra context sent to the LLM for scenario generation
  • Project assets — Images, video, audio, fonts (e.g. logos, watermarks, product photos) reusable across works and in video generation
  • Editor templates — Reusable overlay compositions (text + image) that can be applied to any work in the project

Projects are scoped to a user. You list, create, update, and delete projects via the UI or Projects API.

When creating a project, you can choose a project type to start with optimized prompts:

TypeDescription
blankPhotorealistic defaults. Fully customizable.
saasSaaS product demos. UI/UX focused, dashboard walkthroughs.
newsNews/media content. Broadcast journalism, documentary style.
socialSocial media/marketing. Trendy, engaging, brand-focused.
ecommerceE-commerce products. Product showcases, lifestyle shots.
educationalTutorials. Clear instruction, step-by-step guides.

Each type pre-configures systemPrompt, imageSystemPrompt, videoSystemPrompt, and styleGuide.

A work is one video pipeline inside a project. Each work has:

  • Name, mode (style_transfer or content_remix)
  • Product/subject info — productName, productDescription, targetAudience, language, videoDuration, sceneCount
  • Resolution — Output dimensions for image/video generation and export ({ width, height }). Default: { width: 1080, height: 1920 } (Portrait 9:16)
  • Reference video (optional) — For style transfer; uploaded and analyzed in the pipeline
  • Selected assets — Asset IDs from the project to use in video generation
  • Asset usage mode — How to use selected assets: reference (AI learns style) or direct (use actual images)
  • Analysis — Result of reference video analysis (scenes, tone, colors, tempo)
  • Scenes — Scenario output (image_prompt, video_prompt, duration, transition per scene)
  • Generated scenes — Per-scene image and video URLs and status
  • Editor state — Timeline (video, text, image, audio tracks), overlays, transitions, export settings

One project can have many works. Each work progresses through the same 5-step pipeline (Init → Analyze → Scenario → Generate → Editor).

When a work has selected assets, you can control how they’re used:

ModeBehavior
referenceAI analyzes asset descriptions and generates similar-looking images. Assets should have descriptions (use Vision AI analyze).
directSkip AI image generation. Use actual asset images directly, only generate video (animation). Scene count auto-adjusts to match asset count.
User
└── Project (style guide, prompts, assets, templates)
├── Work 1 (one video)
├── Work 2 (another video)
└── Work 3 …

When you create a work, it inherits the project’s system prompt, analyzer prompt, and image/video instructions. You can override them at the work level. The project style guide is applied in scenario, image, and video generation unless overridden.