3.0.0 — Subtitle & Narrator Independent Control + AI Subtitle Styling + Digital Anchor + Image Generation
Subtitle & Narrator Independent Control + AI Subtitle Styling + Digital Anchor + Image Generation
What's New
Features & Improvements
- Subtitle & narrator independent control —
SubtitleConfigelevated to a peer ofAudioConfig, each with its ownenabledtoggle; four combination modes (narrator+subtitle, narrator-only, subtitle-only, silent); newsubtitle_enabledparameter 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_hintsaccepts 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 sidecarsubtitle_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_subtitlereplaced by independent_step_audio+_step_subtitlein creative and manuscript pipelines. - Subtitle rendering improvements — multi-line display with 0.3s overlap, 0.8s transition overlap, two-pass
extend-endtiming, safe-margin overflow protection, LLM vertical-zone positioning for visual variety.
Bug Fixes
subprocess.runnow passesstdin=DEVNULL, preventing background-process SIGTTIN hangs.SilentTTSEngine.generate()returnsNone→ emptydict, 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.
About Agnes Video Generator
A completely free, open-source AI video generator: turn text into multi-scene AI videos with narration and subtitles.
- Free and open source — no payment, no high-end GPU required
- Text-to-video, image-to-video, digital anchor and image generation
- Powered by free Agnes AI models, rendered entirely in the cloud
- Runs via Web UI, Docker or npm, and supports self-hosting
Related Pages
Release Notes
Changelog for Agnes Video Generator: new features, improvements and bug fixes for every release.
Try Online Demo | Agnes Video
Enter a prompt and generate AI videos and images instantly
Text to Video API — Free AI Video Generation API, No GPU | Agnes Video
Programmatically access Agnes AI's free video models. Text-to-video, image-to-video, and narration APIs. Developer-friendly, zero-cost, no high-end GPU required.
Why you can trust this
Written by a practicing developer and cross-checked against authoritative primary sources such as arXiv papers, vendor technical reports, and the Stanford AI Index.
SandGrid@lcy362
Author of Agnes Video Generator · Full-stack Developer
Independent developer and author of Agnes Video Generator, an open-source (MIT) AI video generation tool built on Agnes AI's free video models. Has helped 1,000+ creators produce AI videos at zero cost. Focused on making video generation models accessible and production-ready; content is based on hands-on practice and primary research.
View on GitHubLast updated：2026-08-21
Ready to Start Creating?
"Making world-class AI belong to everyone." — Bruce Yang. It's completely free, no credit card, and you won't need a high-end GPU. Your first AI video starts at zero cost. Want to use Agnes AI's free video models? This is the easiest way in.
Clone the GitHub repo and launch in 2 minutes