Quick Start

How it works

1. Write or import a Fountain scene
2. Click Build RenderPackage to create the draft
3. Use PromptTuner to review and approve prompts
4. Click Export RenderPackage to download the zip
5. Use the package manually or with a capable agent
6. Generate takes in your AI-video workflow
7. Pick keepers and edit the scene

Write scene

Write or export a Fountain screenplay scene.

Fountain format

Build RenderPackage

Upload your screenplay, select a scene, then click Build RenderPackage. RenderScript creates a package draft first so you can check the prompts before export. For full scripts, turn on Project Bundle mode before export to create one linked scene package per scene plus a project manifest. The source .fountain file is included in the package for reference. No video is generated.

Use PromptTuner

Review and edit reference prompts and shot prompts. Reference prompts are drafts for images, voices, style, locations, props, and other reusable production details. Approve a reference only when you want agents and video tools to treat it as something that should stay consistent. Prompt Assist is optional wording help for the current card only; manual editing is always enough.

PromptTuner tips

• Approve a reference only when it is ready to reuse across shots.
• Use Must stay the same for details like face, costume, prop shape, colour, scale, lighting, or accent.
• Use Do not change for common drift, such as changing a coat, adding extra props, modernising a location, or changing voice style.
• Leave uncertain fields blank. A draft prompt can still export without becoming an approved reference.
• Use Prompt Assist only when you want help rewriting the wording. It changes one card at a time and does not approve anything for you.

Export RenderPackage

Click Export RenderPackage to download one RenderPackage.zip. Start with RENDERPACKAGE.pdf; it is the storyboard-led shooting guide. The creator files sit at the root, and the agent handoff files live inside DEVELOPER_FILES/.

Use across a larger project

Each RenderPackage covers one scene. A project bundle wraps multiple scene packages with project_manifest.json, project_index.json, scene batches, and shared project_refs/. Reuse the same approved character, location, prop, style, or voice references across them.

Use it manually or with agent-assisted workflows

For manual work, follow RENDERPACKAGE.pdf, open COPY_PASTE_PROMPTS.docx, and copy the prompts into your AI video tool. For agent-assisted work, hand over the whole zip; agents and local tools can read DEVELOPER_FILES/rpack.json, AGENT_ORCHESTRATION.md, and the prompt packs.

Approval checkpoints

RenderScript keeps the human approvals simple: approve prompts before export, approve visual or voice references before treating them as reusable, review takes, pick keepers, and decide what needs revision.

After export, choose the image or video tool you want to use. If an agent is helping, tell it whether you will run that tool yourself or whether it has permission to operate the tool for you. It may ask for access through your normal account, browser session, workspace, or API setup. It may also ask where to upload references, what aspect ratio or duration to use, how many takes to generate, and when it should stop for approval before spending credits.

Generate takes

Generate outside RenderScript in your chosen AI-video tool or workflow. Start with one test take. If close, generate 2–3 more.

Pick keepers

Use KEEPER_SHEET.csv and DEVELOPER_FILES/keeper_decisions.csv to track useful takes.

Edit the scene

Use KEEPER_SHEET.csv, then cut the sequence and finish audio in post-production.

RenderScript does not generate video. RenderPackage includes no generated clips. It assists downstream AI video generation with AI video models.

For Filmmakers

AI video generation gets messy fast. Prompts drift, characters change, props disappear, and every shot can become a new negotiation with the machine.

RenderScript gives your screenplay scene a shooting pack before generation begins.

It helps you:

• See the scene as shot cards
• Tune reference prompts and shot prompts before export
• Attach approved reference images or voice samples you already have so the package carries them forward
• Export one package that contains prompts, refs, keeper tracking, and agent handoff files
• Use the package manually or pass the zip to a capable agent
• Generate test takes
• Pick keepers
• Cut picture and finish audio in post-production

Workflow:

Write the scene.
Build a RenderPackage draft.
Approve prompts in PromptTuner.
Export the RenderPackage zip.
Use it manually or with an agent-assisted workflow.
Generate takes outside RenderScript.
Pick keepers in the keeper sheet.
Cut the sequence.
Finish audio in post-production.

Agent handover

The handover is the exported RenderPackage itself. Give the zip to your chosen agent or workflow. The agent-readable instructions live inside DEVELOPER_FILES/AGENT_ORCHESTRATION.md, with the shot plan, prompt packs, approval checkpoints, and logging templates beside it.

After export, choose the image or video tool you want to use. If an agent is helping, tell it whether you will run that tool yourself or whether it has permission to operate the tool for you. It may ask for access through your normal account, browser session, workspace, or API setup. It may also ask where to upload references, what aspect ratio or duration to use, how many takes to generate, and when it should stop for approval before spending credits.

Supported handoff routes

Manual workflow Agent handoff Approval gated

RenderScript does not replace direction or editing. It gives creators a clearer way to move from screenplay scene to downstream AI video generation with AI video models.

Project continuity

Each RenderPackage covers one scene, but a film is more than one scene. For a larger project, turn on Project Bundle mode before export. The bundle keeps scene packages together, records the project manifest, and gives agents shared project refs for style, characters, locations, props, voice, and keeper decisions.

When you already have approved reference images or voice samples, add them in PromptTuner beside the matching reference prompt. They export with the package so the next RenderPackage can reuse the same approved references.

About RenderScript

RenderScript prepares screenplay scenes for AI video production.

It converts one scene into a storyboard-led RenderPackage: shot cards, base prompts, reference folders, keeper tracking, post-production notes, and developer files for downstream integrations.

Larger projects can use a Project Bundle: one RenderPackage per scene, with project_manifest.json, project_index.json, scene batches, and shared project_refs/ so agents and local tools can read them as related scene packages.

Persistence comes from the Project Bundle and package data: shared reference images, voice notes, stable reference folders, shot bindings, keeper logs, project_manifest.json, and clear project refs. RenderScript records approved references; agents, local tools, or the manual workflow use them.

PromptTuner lets creators edit prompts, add source assets, and approve which references should stay consistent before export. What you see there is what goes into the RenderPackage.

RenderPackage includes no generated video. It is assistance for downstream AI video generation with AI video models.

Agent orchestration

Capable agents can follow the package. You stay the director.

• Agents can read the package.
• Agents can follow the shot plan.
• Agents can use prompt packs.
• Agents can log takes.
• Humans approve prompts, visual and voice references, keepers, and revisions.

RenderPackage dissection

• RENDERPACKAGE.pdf — storyboard-led shooting guide
• COPY_PASTE_PROMPTS.docx — reference and shot prompts
• KEEPER_SHEET.csv — take and keeper tracking
• refs/ — reference folders
• generated_shots/ — working folders for your takes and keepers
• DEVELOPER_FILES/ — agent and developer-readable package layer
• DEVELOPER_FILES/rpack.json — machine-readable package contract
• DEVELOPER_FILES/AGENT_ORCHESTRATION.md — agent handoff guide
• DEVELOPER_FILES/action_plan.json — shot execution plan
• DEVELOPER_FILES/approval_checkpoints.json — human approval checkpoints
• DEVELOPER_FILES/take_log.csv and DEVELOPER_FILES/keeper_decisions.csv — logs for takes and keeper choices

Developer layer

The developer layer includes rpack.json, shot_list.csv, bindings.csv, prompt packs, orchestration files, and logs. It is there for agents, local tools, and developer workflows; it does not dominate the creator path.

Across a Project Bundle, project_manifest.json, project_index.json, project_refs/, each scene's DEVELOPER_FILES/rpack.json, bindings.csv, reference folders, take_log.csv, and keeper_decisions.csv give agents and local tools structured continuity context.

RenderScript is an agent-actionable AI-video workflow layer, not a platform-specific generator.

RenderScript does not generate video. It builds the shooting pack you use with your preferred AI video tools.

For Developers

RenderPackage's place is as an inspectable production package for safe handoff. It gives developers, agents, and local tools structured scene data, approved references, prompts, shot IDs, keeper logs, and approval checkpoints without asking anyone to trust a black-box skill.

Creators use the Studio UI and the creator-facing package files. Developers can build QA checks, adapters, and workflow bridges around the open package structure. The repo also includes an optional OpenClaw handoff skill for teams that use it.

DEVELOPER_FILES contains rpack.json, provenance.json, shot_list.csv, bindings.csv, AGENT_ORCHESTRATION.md, action_plan.json, approval_checkpoints.json, provider_capabilities.example.json, package_map.md, and prompt packs.

Prompt edits from PromptTuner are written into DEVELOPER_FILES/rpack.json, provenance.json, and generated prompt packs.

For multi-scene projects, pass the Project Bundle to your chosen agent or workflow. Agents can read project_manifest.json, follow scene batches, and compare each package's rpack.json, bindings.csv, reference folders, prompt packs, take_log.csv, and keeper_decisions.csv to maintain continuity. This is persistent project context supplied through files, not automatic execution inside RenderScript.

Optional OpenClaw skill

The open-core repo includes the text-first skill at skills/renderscript-openclaw-handoff/. It is a plain instruction folder, not a bundled executable. Copy it into ~/.openclaw/workspace/skills/renderscript-openclaw-handoff/, inspect the instructions, then run openclaw skills check. The skill is an optional example of how an agent can consume the package: no hidden code, no credential handling, and no uploads or generation without creator approval.

Canonical source: RenderScript OpenClaw skill on GitHub.

Handoff contract

• AGENT_ORCHESTRATION.md explains how agents and local tools can consume the package.
• rpack.json exposes shot IDs, shot intent, base prompts, reference requirements, reference folders, prompt pack paths, and provenance paths.
• action_plan.json and approval_checkpoints.json keep the package agent-actionable, not auto-executable.
• provider_capabilities.example.json is a template for optional workflow capability hints.

What developers can build

• Agent workflows that prepare safe handoff batches
• Execution templates for serious AI video workflows that accept useful references
• Validation and QA checks around screenplay, package outputs, and generated video takes
• Internal tools that wrap RenderPackage generation
• Pipeline integrations for studio or cloud workflows
• Downstream orchestration systems that log generation, video QA, keeper review, and provenance capture

Generation execution, account access, billing, media upload, render queues, and automatic keeper choice remain outside RenderScript.

The machine-readable package contract lives in:

DEVELOPER_FILES/rpack.json

CLI quick start

• python3 -m pip install -e .
• renderscript package path/to/script.fountain --scene 1 --provider universal -o ./out/package.zip

Where to go next

• Studio UI
  renderscript.studio
• GitHub
  https://github.com/ioioioan/renderscript