Video AI Repos, 2026 (Prompt Library)

Alibaba's Apache-licensed text-to-video. The 1.3B variant runs on a mid-range gaming GPU.

Tiered model sizes (1.3B → 14B). The 1.3B fits in 8 GB VRAM and is the gentlest entry to local video gen. Multilingual, supports image-to-video and video-to-audio.

Mixture-of-Experts upgrade — separate high-noise and low-noise experts handle layout vs detail.

Currently the gold standard for open-source quality. Worth the bigger VRAM ask if you have a 24 GB+ card.

Tencent's 13B unified text-to-video, image-to-video, video-to-video model.

Best-in-class temporal consistency past 10 seconds. Strong character/face stability and prompt-driven camera moves. Quantized v1.5 with offload runs on 14 GB.

Lightricks' speed-tier model. Multiscale rendering — fast low-res motion pass, then refined.

Real-time-ish generation. 4-second clips at 720p in under 30 seconds on a 4090. The pick when you want to iterate fast.

10B Asymmetric Diffusion Transformer with a custom VAE that compresses video 128x.

Best-in-class motion realism — water, fabric, jitterless camera. Apache 2.0, so safe for commercial pipelines.

5B model from Zhipu AI focused on prompt adherence and reproducibility.

The most researcher-friendly option. Solid Diffusers support, quantizes cleanly to 8-bit. Strong for cinematic story-driven content.

HunyuanVideo fine-tuned on 10M+ film and TV clips for human-centric scenes.

If your output needs realistic people, this is the pick. Cinematic-quality faces and motion.

Node-based GUI/backend for diffusion models. The de facto orchestration layer.

Every new video model — Wan, Hunyuan, LTX, Mochi — ships with a ComfyUI workflow file as standard. Used by Netflix, Apple, Ubisoft. Raised $30M at $500M valuation in April 2026.

HeyGen's open-source HTML-to-video framework. Compose videos by writing HTML, CSS, and JS — no React, no DSL.

Explicitly agent-first. LLMs already know HTML inside-out, which means Claude Code and Cursor write compositions natively. Deterministic renders, swappable animation runtimes (GSAP, Lottie, Three.js). Inspired by Remotion but lighter to author.

OpenAI's reference speech-to-text. The default for serious transcription.

Foundation of nearly every video-to-LLM pipeline. Outperforms human transcribers in most conditions except heavy background noise.

C/C++ port of Whisper. Runs on Apple Silicon, Raspberry Pi, anywhere — no Python.

Embeds transcription into desktop and mobile apps. The pick when Python isn't an option.

Downloads video and audio from YouTube, TikTok, Instagram, Twitter, ~1000 other sites.

First step of every 'make AI summarize this video' pipeline. Pair with Whisper for an LLM-ready transcript from any URL.

100% local transcription + subtitling suite with a web UI. Powered by Faster-Whisper.

Self-hosted alternative to Otter.ai or Rev. Pulls from any yt-dlp-supported URL, edits subtitles, translates via LibreTranslate. Offline.

Transcribe + summarize videos through any OpenAI-compatible API.

Subtitle-first architecture — uses native YouTube subtitles when present (instant) and falls back to Whisper. Much faster than naive pipelines.

Edit videos by chatting with Claude Code. Drop raw footage in a folder, get a finished mp4.

Cuts filler words and dead space, auto color-grades, burns subtitles, generates animation overlays via HyperFrames / Remotion / Manim, and self-evaluates each cut for jumps and audio pops. Agent-native end-to-end editing — no presets, no menus.

Open-source desktop app that turns novels and scripts into animated short dramas.

Integrates AI scriptwriting, storyboarding, character generation, and video gen into one cross-platform GUI. Aimed at indie creators batching content cheaply.

Lip-sync any audio to any face video.

Foundation tech behind half the AI dubbing startups. Pair with Whisper + GPT translation + ElevenLabs voice cloning for full multilingual pipelines.

Modern lip-sync that handles head pose and lighting better than Wav2Lip.

The drop-in upgrade when Wav2Lip outputs look too obviously synced. Better quality on natural-talking footage.

Adds motion to Stable Diffusion outputs.

Plugs into existing SD workflows — your LoRAs and checkpoints still work. Cheapest way to get animation out of an image-gen stack you already run.

Image-to-video from Stability AI. Older but still in active use.

When you need a known-quantity image-to-video baseline. Mature ComfyUI integrations.