v4.6.0

Release v4.6.0 — Poetry Video Task Type + MultiScene Pipeline Refactor

Release date: 2026-07-12

Overview

v4.6.0 is a minor version that introduces the **Poetry Video ** task type — ancient poems are split into scenes by the LLM, each line is read aloud by TTS and aligned to video subtitles — and refactors the long-video pipelines onto a shared MultiScenePipeline base.

Usage

From v4.5.0:

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

What's New

Features & Improvements

• Poetry Video task type — new POST /api/tasks/poetry: the LLM splits a poem into scenes (original line | visual description format), each line is rendered by TTS narration, scenes are generated as t2v clips, and subtitles are time-aligned to the narration and concatenated.
• Poetry scene config aligned with creative video — per-scene duration model and duration-extraction mode; scene count / watermark / narration speed options.
• Two-phase poetry generation — scenes are submitted in a batch and awaited in parallel for faster completion.

Refactoring & Optimizations

• MultiScenePipeline base class — creative / manuscript / anchor / poetry long-video pipelines now share one template-method base (build_scenes → build_reference_images → generate_videos → audio+subtitle → composite), eliminating duplicated per-pipeline logic.
• Poetry generation reuses the base generator — video generation is hoisted to MultiScenePipeline._generate_videos.

Bug Fixes

• Per-scene TTS audio truncation in poetry videos (only the first half audible).
• Resume reused a stale final_clip.mp4 without an audio track.
• Poetry progress step indicator not displayed.
• switchTaskType missing a closing brace, which disabled all page JS.