Open-source AI podcast clipper. Generate vertical clips with face tracking and burned-in captions. CLI, MCP server, web app.
Python ai-toolsauto-captionscaptionscliffmpegmcpmcp-serveropusclip-alternativepodcast-clipspodcastsshortform-videovertical-videowhisper
"""Project IR -> FCPXML 1.10 for DaVinci Resolve 20.x.
Resolve quirks the emitter routes around:
- FCPXML transform keyframes mis-translate values (center-origin math). Reframe
is pre-baked into the V1 source layer instead of emitted as keyframes.
- Composite-mode / blend-mode attributes are silently dropped. Alpha is carried
by the asset itself (ProRes 4444); Resolve auto-detects and composites normal.
- PNG image sequences import as one-frame-per-clip. Use single video files.
- Avoid any Studio-only effect node: Resolve 20.2 watermarks the timeline
if it sees one, even if unused.
- Pin to FCPXML 1.10; 1.11+ have intermittent import failures.
"""
from __future__ import annotations
import xml.etree.ElementTree as ET
from fractions import Fraction
from pathlib import Path
from .._shared import fcpxml as fx
from .._shared.timeline_ir import Project
def emit(project: Project, out_path: Path) -> Path:
fmt_id = "r1"
resources: list[ET.Element] = [
fx.make_format(fmt_id, project.fps, project.width, project.height),
]
compounds: list[tuple[str, str, Fraction]] = []
next_asset = 2
next_media = 1000
for short in project.shorts:
src_id = f"r{next_asset}"
next_asset += 1
resources.append(fx.make_asset(
asset_id=src_id,
name=f"{short.title}: source",
media_path=short.source.path,
frames=short.source.duration_frames,
fps=short.source.fps,
format_id=fmt_id,
has_video=True,
has_audio=short.source.has_audio,
audio_channels=short.source.audio_channels,
))
v2: tuple[str, Fraction] | None = None
if short.captions:
cid = f"r{next_asset}"
next_asset += 1
resources.append(fx.make_asset(
asset_id=cid,
name=f"{short.title}: captions",
media_path=short.captions.path,
frames=short.captions.duration_frames,
fps=short.captions.fps,
format_id=fmt_id,
has_video=True,
has_audio=False,
))
v2 = (cid, fx.frames_to_seconds(short.captions.duration_frames, short.captions.fps))
v3: tuple[str, Fraction] | None = None
if short.logo:
lid = f"r{next_asset}"
next_asset += 1
resources.append(fx.make_asset(
asset_id=lid,
name=f"{short.title}: logo",
media_path=short.logo.path,
frames=short.logo.duration_frames,
fps=short.logo.fps,
format_id=fmt_id,
has_video=True,
has_audio=False,
))
v3 = (lid, fx.frames_to_seconds(short.logo.duration_frames, short.logo.fps))
cmpd_id = f"r{next_media}"
next_media += 1
source_duration = fx.frames_to_seconds(short.source.duration_frames, short.source.fps)
compounds.append((cmpd_id, short.title, source_duration))
resources.append(fx.make_compound_media(
media_id=cmpd_id,
name=short.title,
format_id=fmt_id,
fps=project.fps,
source_duration=source_duration,
v1_asset_id=src_id,
v1_has_audio=short.source.has_audio,
v2=v2,
v3=v3,
))
library = fx.make_project_library(
project_name=project.name,
event_name="podcli",
format_id=fmt_id,
fps=project.fps,
compounds=compounds,
)
fx.write_fcpxml(out_path, resources, library, version="1.10")
return out_path