Skip to content

v3.0.0 — 字幕与旁白独立控制 + AI 智能字幕 + 数字人口播 + 图片生成

Choose a tag to compare

View all tags
@lcy362 lcy362 released this 21 Jun 14:39
· 307 commits to master since this release
v3.0.0
a77d249

Release v3.0.0 — Subtitle & Narrator Independent Control + AI Subtitle Styling + Digital Anchor + Image Generation

Release date: 2026-06-21

Overview

v3.0.0 is a major version that introduces four significant capabilities: independent subtitle/narrator control, AI-powered subtitle styling, a new Digital Anchor video task type, and simple image generation. It also delivers extensive subtitle rendering improvements and system-prompt support for both video and image tasks.

Usage

From v2.2:

git pull
.venv/bin/pip install -r requirements.txt
./start.sh

Breaking change: AudioConfig.subtitle_style has been removed. Legacy task states are auto-migrated on load — no manual intervention required.

What's New

Features & Improvements

  • Subtitle & narrator independent control — SubtitleConfig elevated to a peer of AudioConfig, each with its own enabled toggle; four combination modes (narrator+subtitle, narrator-only, subtitle-only, silent); new subtitle_enabled parameter on creative and manuscript endpoints.
  • AI-powered subtitle styling (LLM mode) — style_mode "fixed" (global style) or "llm" (per-subtitle position/color/font size decided by the LLM); style_hints accepts natural-language guidance ("key lines in red, summaries in yellow"); generate_subtitle_styles() sends all subtitles to the LLM in one call and renders via sidecar subtitle_styles.json.
  • Digital Anchor task type — AI-generated presenter via t2i or i2i (user-uploaded reference photo); manuscript split into 5–12s paragraphs, each generating a unique i2v clip with different gestures/expressions; unified TTS + LLM-optimized subtitle overlay; audio from the model or post-concatenation; POST /api/tasks/anchor.
  • Simple image generation — new fifth tab; lightweight SimpleImageTask; system-prompt support; GET /api/image/{id} to retrieve results.
  • System prompt support — optional system prompt for simple video and image tasks, prepended to the final prompt.

Refactoring & Optimizations

  • Pipeline step split — _step_audio_subtitle replaced by independent _step_audio + _step_subtitle in creative and manuscript pipelines.
  • Subtitle rendering improvements — multi-line display with 0.3s overlap, 0.8s transition overlap, two-pass extend-end timing, safe-margin overflow protection, LLM vertical-zone positioning for visual variety.

Bug Fixes

  • subprocess.run now passes stdin=DEVNULL, preventing background-process SIGTTIN hangs.
  • SilentTTSEngine.generate() returns None → empty dict, fixing SRT generation in subtitle-only mode.
  • SRT timeline is based on actual audio duration instead of estimated video duration.
  • Image generation failures now surface the specific error (HTTP status + body + traceback).
  • Defensive UTF-8 encoding fix in the digital-anchor pipeline; TTS now runs first to obtain the real audio duration.
Assets 2
Loading