Real2Motrix/scripts/render_launch_video.py
eust-w d13e06907b
Some checks are pending
Core checks / backend (push) Waiting to run
Core checks / frontend (push) Waiting to run
🚀 feat: prepare auditable open-source alpha and launch media
2026-10-26 17:38:24 +08:00

203 lines
11 KiB
Python
Raw Permalink Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

"""Compose a narrated launch explainer from the real, locally recorded UI.
Requires macOS say, FFmpeg/FFprobe and Pillow. No private capture footage or
third-party music is used. Pass a CJK font file with --font on other setups.
"""
import argparse
import json
import re
import subprocess
from pathlib import Path
from PIL import Image, ImageDraw, ImageFont, ImageOps
SCENES = [
("3D 模型进仿真,还差哪一步?", "VISUAL / COLLISION / STRUCTURE",
"把三维模型带进机器人仿真,不只是能看。还需要碰撞几何、场景结构,以及可检查的输出。", "motion"),
("Real2Motrix", "本地采集与资产处理工作台 / ALPHA",
"这是 Real2Motrix,一个开源的本地采集与资产处理工作台。目前处于 Alpha 阶段。", "desktop"),
("真实处理,不伪装成扫描", "原创几何测试资产 / 非实拍重建",
"你现在看到的是原创几何测试资产,不是实拍重建。它通过真实后端处理,在桌面端交互预览。", "motion"),
("不止一个模型文件", "MJCF / URDF / MESH / METADATA",
"视觉网格与碰撞网格分别管理,输出 MJCF、URDF 和元数据。导出包还会进行结构校验。", "package"),
("手机采集:实验能力", "轨迹覆盖率 ≠ 表面完整度",
"项目提供手机原生采集和在线上传代码。但真机效果还需要验收,轨迹覆盖率也不等于扫描完整度。", "boundary"),
("把下一步,变成可验收的结果", "RECONSTRUCTION / COVERAGE / SIMULATION",
"接下来重点是表面覆盖反馈、物体和场景分流,以及真实数据集上的重建与仿真验证。", "roadmap"),
("从可复现,走向可信赖", "github.com/eust-w/Real2Motrix",
"代码已经开放。如果你在做机器人仿真、三维重建或移动采集,欢迎运行示例,用真实测试一起改进。", "motion"),
]
def run(command):
subprocess.run([str(value) for value in command], check=True, stdout=subprocess.DEVNULL)
def duration(path):
return float(subprocess.check_output([
"ffprobe", "-v", "error", "-show_entries", "format=duration", "-of", "csv=p=0", str(path)
], text=True).strip())
def wrap(draw, text, font, width):
lines, line = [], ""
for char in text:
if line and draw.textlength(line + char, font=font) > width:
lines.append(line)
line = ""
line += char
if line:
lines.append(line)
return lines
def timestamp(seconds):
millis = round(seconds * 1000)
return f"{millis // 3600000:02}:{millis // 60000 % 60:02}:{millis // 1000 % 60:02},{millis % 1000:03}"
def main():
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--output", type=Path, default=Path("artifacts/launch"))
parser.add_argument("--font", type=Path, default=Path("/System/Library/Fonts/Hiragino Sans GB.ttc"))
parser.add_argument("--voice", default="Tingting")
parser.add_argument("--orbit", type=Path, help="Optional already-trimmed fullscreen orbit recording")
args = parser.parse_args()
out = args.output.resolve()
build = out / "render"
build.mkdir(parents=True, exist_ok=True)
evidence = json.loads((out / "evidence.json").read_text())
if not evidence["validation"]["valid"]:
raise RuntimeError("Do not render a successful-demo claim without validation evidence")
audio_durations = []
subtitles = []
elapsed = 0
for index, (_, _, narration, _) in enumerate(SCENES):
voice = build / f"voice-{index}.aiff"
run(["say", "-v", args.voice, "-r", "205", "-o", voice, narration])
spoken = duration(voice)
audio_durations.append(spoken + 0.65)
chunks = [part for part in re.split(r"[。!?]", narration) if part]
weights = [len(part) for part in chunks]
cursor = 0
for part, weight in zip(chunks, weights):
end = cursor + spoken * weight / sum(weights)
subtitles.append((elapsed + cursor, elapsed + end, part))
cursor = end
elapsed += spoken + 0.65
(out / "Real2Motrix.zh-CN.srt").write_text("\n\n".join(
f"{i + 1}\n{timestamp(start)} --> {timestamp(end)}\n{text}"
for i, (start, end, text) in enumerate(subtitles)
) + "\n", encoding="utf-8")
# The raw capture includes setup. Only the fullscreen orbit is distributed.
motion = args.orbit.resolve() if args.orbit else build / "orbit.mp4"
if not args.orbit:
run(["ffmpeg", "-y", "-v", "error", "-ss", "8", "-i", out / "desktop-recording.webm",
"-t", "10", "-an", "-c:v", "libx264", "-crf", "18", "-pix_fmt", "yuv420p", motion])
font = lambda size: ImageFont.truetype(str(args.font), size)
bg, fg, muted, teal, amber = "#171b1c", "#f4f6f5", "#acb8b7", "#53d5bc", "#ffce71"
for orientation, (width, height) in {"landscape": (1920, 1080), "portrait": (1080, 1920)}.items():
vertical = orientation == "portrait"
margin = 64
media = (64, 490, 952, 720) if vertical else (64, 275, 1792, 640)
chapter_start = 0
clips = []
for index, (title, subhead, narration, kind) in enumerate(SCENES):
seconds = audio_durations[index]
plate = Image.new("RGB", (width, height), bg)
draw = ImageDraw.Draw(plate)
draw.text((margin, 50), "REAL2MOTRIX / OPEN-SOURCE ALPHA", font=font(25), fill=teal)
title_lines = wrap(draw, title, font(64 if vertical else 62), width - margin * 2)
y = 130
for line in title_lines:
draw.text((margin, y), line, font=font(64 if vertical else 62), fill=fg)
y += 90
draw.text((margin, y + 15), subhead, font=font(27), fill=muted)
mx, my, mw, mh = media
if kind in {"motion", "desktop"}:
screenshot = Image.open(out / ("desktop.png" if kind == "desktop" else "scene.png"))
if kind == "motion":
screenshot = screenshot.crop((330, 260, 1590, 1000))
screenshot = ImageOps.contain(screenshot, (mw, mh))
plate.paste(screenshot, (mx + (mw - screenshot.width) // 2, my + (mh - screenshot.height) // 2))
else:
if kind == "package":
lines = [("01", "scene.xml / scene.urdf"), ("02", "视觉网格 + 碰撞网格"),
("03", "物体清单 + 重建元数据"), ("04", "结构校验:通过")]
elif kind == "boundary":
lines = [("有源码", "iOS / ARKit 原生采集"), ("有接口", "帧与网格的在线上传"),
("待验收", "真机采集与补扫效果"), ("需区分", "轨迹覆盖 ≠ 表面覆盖")]
else:
lines = [("01", "真实采集基准与失败案例"), ("02", "场景 / 物体分别处理"),
("03", "表面覆盖与不确定性反馈"), ("04", "目标仿真器内的物理验证")]
for row, (label, value) in enumerate(lines):
row_y = my + 35 + row * 145
draw.line((mx, row_y - 15, mx + mw, row_y - 15), fill="#3b4546", width=2)
draw.text((mx, row_y), label, font=font(27), fill=teal)
draw.text((mx if vertical else mx + 170, row_y + 49 if vertical else row_y), value,
font=font(38 if vertical else 46), fill=fg)
draw = ImageDraw.Draw(plate)
label_y = 1265 if vertical else 934
draw.text((margin, label_y), "原创几何测试资产 · 非实拍重建", font=font(26), fill=amber)
if vertical:
draw.text((margin, 1340), "本地处理 / 可检查输出 / 实验性手机采集", font=font(25), fill=muted)
draw.text((margin, height - 38), f"{index + 1:02} / 07 真实软件录屏 + 合成配音", font=font(19), fill=muted)
image_path = build / f"{orientation}-{index}.png"
plate.save(image_path)
if index == 0:
plate.save(out / f"cover-{orientation}.png")
inputs = ["ffmpeg", "-y", "-v", "error", "-loop", "1", "-framerate", "30", "-i", image_path]
filters = []
current = "0:v"
next_input = 1
if kind == "motion":
inputs.extend(["-stream_loop", "-1", "-i", motion])
filters.append(f"[1:v]crop=1260:740:330:260,scale={mw}:{mh}:force_original_aspect_ratio=decrease,pad={mw}:{mh}:(ow-iw)/2:(oh-ih)/2:color=0x171b1c[screen]")
filters.append(f"[0:v][screen]overlay={mx}:{my}[media]")
current = "media"
next_input += 1
for caption_index, (start, end, text) in enumerate(subtitles):
if start < chapter_start - 0.01 or start >= chapter_start + seconds:
continue
subtitle = Image.new("RGBA", (width, 180), (0, 0, 0, 0))
painter = ImageDraw.Draw(subtitle)
lines = wrap(painter, text, font(40 if vertical else 38), width - 128)
for line_index, line in enumerate(lines):
x = (width - painter.textlength(line, font=font(40 if vertical else 38))) // 2
painter.text((x, line_index * 58), line, font=font(40 if vertical else 38), fill=fg)
subtitle_path = build / f"caption-{orientation}-{caption_index}.png"
subtitle.save(subtitle_path)
inputs.extend(["-loop", "1", "-framerate", "30", "-i", subtitle_path])
label = f"sub{caption_index}"
filters.append(f"[{current}][{next_input}:v]overlay=0:{1510 if vertical else 986}:enable='between(t,{start-chapter_start:.3f},{end-chapter_start:.3f})'[{label}]")
current = label
next_input += 1
inputs.extend(["-i", build / f"voice-{index}.aiff"])
filters.append(f"[{current}]fade=t=in:st=0:d=0.22,fade=t=out:st={seconds-.22:.3f}:d=0.22,format=yuv420p[video]")
clip = build / f"{orientation}-{index}.mp4"
run(inputs + ["-filter_complex", ";".join(filters), "-map", "[video]", "-map", f"{next_input}:a",
"-af", "apad,loudnorm=I=-16:TP=-1.5:LRA=11", "-t", f"{seconds:.3f}",
"-r", "30", "-c:v", "libx264", "-preset", "veryfast", "-crf", "19",
"-threads", "4", "-c:a", "aac", "-b:a", "192k", "-ar", "48000", clip])
clips.append(clip)
chapter_start += seconds
concat = build / f"{orientation}.txt"
concat.write_text("\n".join(f"file '{p}'" for p in clips) + "\n")
target = out / f"Real2Motrix-intro-{orientation}-1080p.mp4"
run(["ffmpeg", "-y", "-v", "error", "-f", "concat", "-safe", "0", "-i", concat,
"-c", "copy", "-movflags", "+faststart", target])
print(f"Rendered {target.name}: {duration(target):.2f}s", flush=True)
(out / "media-provenance.json").write_text(json.dumps({
"source": "real application recording of original synthetic geometry",
"not_demonstrated": ["real phone scan", "video reconstruction fidelity", "physics runtime", "3DGS"],
"voice": f"macOS say {args.voice}; synthetic narration", "music": "none",
"narration": [scene[2] for scene in SCENES], "package_sha256": evidence["package_sha256"],
}, ensure_ascii=False, indent=2), encoding="utf-8")
if __name__ == "__main__":
main()