Available Pipelines

OpenMontage provides 12 pipelines as declarative YAML manifests in pipeline_defs/. Each manifest defines the stage sequence, tools available to the agent, produced artifacts, human approval defaults, review focus, and compatible style playbooks. The agent reads the manifest to determine order and then follows the matching stage-director skill in skills/pipelines/.

All manifests are validated against schemas/pipelines/pipeline_manifest.schema.json at load time. Use the pipeline loader to inspect them at runtime:

from lib.pipeline_loader import list_pipelines, load_pipeline

print(list_pipelines())
manifest = load_pipeline("cinematic")
print(manifest["orchestration"])

See Pipeline System for manifest structure and loader helpers, and Running Pipelines for how an agent executes a chosen pipeline.

Pipeline Catalog

Pipeline	Category	Stability	Description
animated-explainer	generated	production	AI-produced explainer with research, narration, visuals, and music.
animation	animation	production	Motion graphics and kinetic typography using Manim, Remotion, or HyperFrames.
avatar-spokesperson	talking_head	production	Avatar-driven presenter videos with focus on lip sync.
character-animation	animation	beta	Local rigged cartoon characters with SVG rigs, pose libraries, GSAP timelines, and HyperFrames rendering.
cinematic	cinematic	production	Trailer, teaser, and mood-driven edits; supports reference video input.
clip-factory	custom	beta	Batch short-form clips extracted from a long source video.
hybrid	hybrid	production	Source footage combined with AI-generated support visuals.
localization-dub	custom	beta	Subtitle, dub, and translate existing video.
podcast-repurpose	hybrid	beta	Podcast highlights turned into video with companion deliverables.
screen-demo	screen_recording	production	Software screen recordings and walkthroughs emphasizing legibility.
talking-head	talking_head	beta	Footage-led speaker videos.
framework-smoke	custom	test	Minimal two-stage pipeline for framework validation.

Standard Stage Progression

Most pipelines follow the canonical flow:

research → proposal → script → scene_plan → assets → edit → compose → publish

Creative stages (research, proposal, script, scene_plan) default to human_approval_default: true. Technical stages (assets, edit, compose) default to automatic progression unless the checkpoint policy or manifest overrides this. See Checkpoint System for status values and Schemas and Artifacts for the canonical artifact per stage.

Stage Variations

• character-animation inserts character_design and rig_plan before scene_plan. It emits a HyperFrames workspace under projects/<name>/hyperframes/ and renders via renders/final.mp4. See Composition Runtimes.
• cinematic and animated-explainer use the v2.0 split of research + proposal (instead of a single legacy idea stage). Both declare reference_input.supported: true with analysis_depth: deep.
• framework-smoke is restricted to the minimal stages needed for smoke testing.
• clip-factory, localization-dub, and podcast-repurpose are specialized for batch or transformation work and may produce multiple deliverables.
• cinematic enforces motion-required gates when delivery_promise.motion_required is set.

Compatible Playbooks and Extensions

Every manifest lists compatible_playbooks (recommended, also_works, and whether custom playbooks are allowed). Playbook selection is locked in the proposal stage. See Style Playbooks.

Manifests also declare extensions flags controlling custom scripts, playbooks, skills, and tools. Use check_extension_permitted and get_permitted_extensions from the loader to validate before use.

For full stage order including optional sub-stages, call get_stage_order(manifest, include_sub_stages=True). For the union of tools required across a pipeline, call get_required_tools(manifest).