Tools / Poetry Cinema

Poetry to 30s vertical micro-cinema (9:16). FULL: create_project → define_persona → break_poem → write_script → generate_storyboard → generate_scenes → add_layers → assemble_film. QUICK: start at break_poem with defaults.

Resume Project

Check what already exists for a project and determine the next step. Agent reads user files via MCP tools and passes contents here. Returns what's ready, what's missing, and the next skill to call with pre-loaded inputs.

Returns: Project state summary with next_skill and pre-loaded next_inputs ready to pass to the next call.

Create Project

Set up a new poetry cinema project with production bible configuration. Free when using scratch cinema defaults; paid when parsing an uploaded production bible (LLM call). Users upload their production bible doc to ToolRouter files first, then pass the file path.

Returns: Project configuration saved to files with visual style and film rules

Define Persona

Define a recurring or per-poem character and generate turnaround sheets (front/left/right/back portraits) for visual consistency across scenes. Uses the persona-generator tool for turnaround sheet creation. Takes ~70s per persona. Async — poll with get_job_result.

Returns: Persona definition with turnaround sheet URLs saved to project files. Render sheets inline via sheet_url.

Break Poem into Scenes

Divide a poem into ~12 scenes with timing, emotional arc (4-beat: hook/setup/disruption/release), and colour phase assignments. Poem text can be passed directly or read from an uploaded file. Saves scenes.json to project files.

Returns: Scene breakdown with timing, emotional beats, and colour phases saved to scenes.json

Write Script

Generate the full storyboard document from the scene breakdown — detailed visual descriptions, camera direction, sound design, and text treatment per scene. Reads scenes.json from project files. Saves script.json. This completes Phase 1 (Pre-production) — user should review before proceeding.

Returns: Storyboard script saved to script.json. User should review before proceeding to storyboard generation.

Generate Storyboard Images

Generate one image per scene using fal.ai with scratch cinema style. Includes persona references for character consistency. Use skip_scenes for user-uploaded images. Completes Phase 2 — review before proceeding.

Returns: Storyboard images saved to project files. User should review before proceeding to video generation.

Generate Scene Video Clips

Image-to-video: animate each storyboard image into a short video clip using fal.ai. Takes 2-5 minutes. Runs asynchronously — returns a job_id, poll with get_job_result until complete. If videos are still generating when the job completes, use check_scenes to poll for them.

Returns: Completed video clips and/or pending video references. Use check_scenes to poll for pending videos.

Check Pending Scene Clips

Poll fal.ai for pending video clips that are still generating. Pass the pending_videos array from generate_scenes output. Returns completed clips and any still-pending references.

Returns: Completed clips saved to project files, plus any still-pending references for further polling

Add Text, Voice & Sound

Composite breathing text overlays, voiceover narration, and ambient sound onto each video clip via Remotion. Voiceover via voice-generator, sound via music-generator. Completes Phase 3 — review before assembly.

Returns: Layered clips with text, voiceover, and sound design. User should review before final assembly.

Assemble Final Film

Stitch all layered clips into the final vertical micro-cinema film. Uses Remotion render server. Adds transitions (brief breath between scenes), holds final frame with last poetry line, fades to black. This completes Phase 4 (Final) — the finished film.

Returns: The finished vertical micro-cinema film. Show film_page to the user for sharing and download.