Features that support LLM workflows.
Lorvara sits between the LLM workflow you already run and the models that render it. 8 layers, and you do not need all of them to start.
01The creation layer
Prompts
A Key Prompt is one picture or one shot, and everything that shapes it.
Describe what you want. The Prompt Agent writes it in the vocabulary your chosen model responds to, the Studio is where you refine it, and the Library keeps every generation and reference beside the prompt that made it.
- Key Prompts: prompt, model, parameters and references, kept together
- Studio: edit and generate in one place
- Library: generations, references and a mood view
02The management layer
Projects
One piece of work, in one place.
A project holds the prompts for a piece of work, the brief they all inherit, and the scenes and shots that order them.
- Prompts grouped by piece of work
- One brief, inherited by every prompt
- Scenes & Shots inside it
03The direction layer
Briefs
Define the cast, places and look once.
Write a brief with your LLM or Lorvara’s Brief Agent, or import one. Use only the blocks you need. Attach it to a project and every prompt inherits it, so the same face, wardrobe, props and grade carry from shot to shot.
- Blocks: Intent, Cast, Locations, Props, Glossary, Look
- Cast: locked reference views per state, plus voice notes
- Locations: an anchor in frame and up to 2 plates
04The shot list
Scenes & Shots
A shot list becomes a queue of prompts.
Paste a treatment or shot list, or build one by hand. Shots are grouped by scene, and each becomes a Key Prompt with its cast, props, location and duration attached.
- Scenes group shots, and a scene can override the brief’s look
- Each shot: what’s in frame, duration, video or still
- Each shot opens in the Studio as a Key Prompt
05The generation system
Models & Credits
Generate in Lorvara, or in the tools you already use.
Image, video, voice, music and sound effects in-app. Copying out, the prompt is rewritten for its destination: aspect syntax, length limit, negatives. Credits are reserved when a job starts and charged for what it used.
- 34 image and video models in-app
- 13 copy-out destinations, each with its own grammar
- Credits: reserved, then charged for what was used
06Node-based workflow
Workspaces
A canvas for the part of the work that is trial and error.
Share references between prompts, run quick test generations, and compare branches side by side. When something works, save it to the prompt that will ship it.
- Node-based canvas for rapid iteration
- References shared between prompts
- Test generations, in place
07The LLM interface
Agents & Skills
Three Claude agents. All of them ask before they act.
The Prompt Agent writes prompts and asks what it can’t infer. The Brief Agent turns a treatment into a brief, or builds one with you. The Skill wizard turns how you work into a Skill. Nothing is generated, saved or spent until you say so.
- Prompt Agent, Brief Agent, Skill wizard
- Skills ship, or you write and import your own
- Styles and camera controls: a brief the agent writes out, not a keyword bolted on
08Collaboration built in
Teams
Everyone on the same brief.
Bring the team into a project: the same brief, scenes and shots, and Key Prompts. Full or viewer seats, and spend reported by person and by media type.
- Full and viewer seats, set by the owner
- One brief, one set of scenes and shots
- Spend reported by person and by media type
Or run it yourself.
Lorvara can be forked and deployed for your company, with your own generation endpoints and your own storage, so nothing has to leave your infrastructure. Custom builds are delivered by Gravital Digital, who build and maintain Lorvara.