Video skills
Video lanes for pi agents in the Fornace team, from the widest concept down to focused tools. All ship in the pi-setup bundle.
Umbrella
The entry gate for all video, image and audio work. Routes every generation, edit, repair and cleanup to current generative models instead of deterministic VFX. Holds the approved editors list (Wan 2.7 edit, FLUX 3 edit) and the dated models file that every other lane reads before choosing a model.
Full production lanes
AI novel and screenplay to short drama series production (DramaFoundry): manuscript ingest, beat-level director planning, character/location identity sheets, automated keyframing, per-beat video generation (H3 Max Turbo, Kling, Wan), character voice casting with AuK and Fish, sound beds with ThinkSound, and automated assembly with checkpoint resume. Shipped in the pi-setup bundle; credentials provisioned automatically via /fornace install or setup.sh. Read the DramaFoundry operator guide ↗ or visit dramafoundry.fornace.net.
Edits clips longer than one model pass: split at scene cuts, edit parts in parallel, retime onto the source frame grid, conform over original audio, verify every hard cut survived. The doctrine for whole footage beauty, object and cleanup passes.
Directs 2 to 3 minute character-led vertical story films assembled from consistent generated scenes, with living voice performances, motivated B-roll, active physical business, populated public scenes, scene-specific sound design, and an adversarial still review gate before rendering.
Motion graphics and animated promos for digital products: kinetic typography, UI showcases, 2.5D camera moves, beat synced cuts, sound design. For UI and data pieces, not filmed footage.
Sound
Generates ambience, foley and sound beds locally on Apple Silicon with the Metal port of ThinkSound. Caption plus chain of thought prompting, deterministic seeds, mixed under voice at minus 33 LUFS.
Burns word highlighted subtitles into talking video with Remotion. ASR timestamps, grammar based phrase splitting, per client brand fonts and highlight colors.
Focused tools
Prompt writing for Kling 3.0 video: text to video, image to video, keyframes, multi-shot sequences, dialogue scenes.
Ideogram V4 via Fal.ai for ad creative: text to image, image to image, LoRA training, logo compositing. Typography heavy ads.
Deterministic heavy lifting only: compression, HEVC transcode, loudness normalization, denoise, lossless remux with FFmpeg and Apple Audio Units. Owns split, retime and conform inside the edit pipeline. Never a visual fix.
Bundle pi-setup skills 1.14.3, updated 2026-09-28.