Visual Pipeline

Status: Approved initial pipeline

Production flow

Lock the episode’s cast, guest, location, style, and audio intent.
Write a 12-panel board using the coverage rules in Little Lanterns Format.
Compile each structured image record through the Anima model adapter.
Generate Anima storyboard candidates locally in ComfyUI.
Review character recognition, location continuity, shot diversity, story clarity, and the motion opportunity in every frame.
Approve one Anima frame per shot. That image becomes the first frame and visual continuity contract for image-to-video.
Write a motion prompt that describes only what changes after the approved first frame.
Animate each approved frame through the local Wan 2.2 14B image-to-video workflow.
Generate the episode’s music, character voices, spoken narration, spoken-word tags, and emotional performance through Suno.
Edit the Wan shots to the approved Suno audio.
Store binaries in SmartGallery and record links and provenance here.

Handoff contracts

Anima → Wan 2.2 14B

Every approved storyboard shot hands off:

shot_handoff:
  shot_id: S01E__-SH__
  approved_first_frame: ""
  anima_prompt_record: ""
  anima_workflow_version: ""
  seed: ""
  dimensions: 864x1152
  visible_characters: []
  continuity_anchors: []
  intended_action: ""
  camera_motion: ""
  environmental_motion: ""
  wan_motion_prompt: ""
  wan_workflow_version: ""

The first-frame image owns appearance, framing, wardrobe, location, props, and starting pose. The Wan motion prompt owns change over time. It must not redesign the frame.

Suno → edit

Every approved audio handoff records:

audio_handoff:
  episode: S01E__
  suno_prompt_record: ""
  music_output: ""
  narration_output: ""
  selected_version: ""
  emotional_arc: ""
  spoken_word_tags: []
  edit_notes: ""

Music and narration may be generated as one integrated performance or as separate approved outputs, but the selected structure must be known before the final picture edit.

Current image baseline

Field	Candidate
Orientation	Portrait video
Aspect ratio	3:4
Working dimensions	864 × 1152
Board size	12 panels
Storyboard / first-frame model	Anima
Image-to-video model	Wan 2.2 14B
Music and narration	Suno
Character strategy	Distinctive visible anchors + structured prompt inheritance
Review strategy	Contact sheet first, individual frames second

Visual acceptance checklist

Lumi and Pip are recognizable without relying on labels.
Close-ups are genuinely close; the face occupies most of the frame.
The sequence includes wide, medium, close-up, ECU, two-shot, OTS, and reverse coverage.
Camera position, subject scale, and composition meaningfully change between panels.
Style phrasing remains unchanged across the sequence.
Location anchors recur without forcing identical backgrounds.
Hands, gaze, props, and screen direction support the intended edit.
No unexplained duplicate character or wardrobe mutation appears.
Every approved frame has a specific, achievable motion plan.
Wan preserves the approved first-frame identity, composition, and location.
Music and narration support the same emotional beats used by the storyboard.

Known lesson

Model quality alone does not produce a storyboard. Shot type must be expressed as camera geometry and subject scale, and each panel must have a distinct editorial job.

Visual-Pipeline