Video Generation
SaaS, open source and skills for video generation work — with the limits of each written down.Before you pick
What decides it
Cost per accepted shot, not price per second. Every rejected attempt bills the same, so control that raises your keep rate decides the bill.
What to avoid
Nothing carries identity between shots, so faces, hair and proportions drift the moment you assemble a sequence rather than judge one clip.
What is changing
Steering is replacing prompting: image, clip and audio references, first and last frames, per-shot direction. Synchronised sound now generates in the same pass.
SaaS
Hosted products you sign up for.Leading
Established picks, at the top of today's aggregate.
- 1Google Veo (Flow)Best realism plus native audio and 4K. Strongest all-rounder for cinematic and narrative scenes.
- 2RunwayThe creator's studio: camera moves, motion brush, reference control and an editor. Safest default.
- 3Kling AIBest value and the community workhorse for TikTok/Reels — photorealistic humans, generous free daily quota.
- 4PikaFast, playful short clips and effects for social. Easy on-ramp for casual creators.
- 5Luma Dream MachineSmooth motion and image-to-video with a simple interface. Solid mid-tier option.
Emerging
Newer challengers, surfaced by two or more sources.
Open source
Repositories you run yourself.Leading
- 1Open-Sora (hpcaitech)The category-defining open-source video generator (~29k GitHub★) — text/image-to-video with performance comparable to HunyuanVideo on VBench.
- 2Wan (Alibaba, Wan2.2)Alibaba's Wan (~16.8k GitHub★) — widely regarded as the strongest open-weight text-to-video model; 1.3B variant runs on 8GB VRAM.
- 3HunyuanVideo (Tencent)Tencent's 13B open video model (~12.4k GitHub★) — a benchmark reference for open-source cinematic text-to-video quality.
- 4CogVideo / CogVideoX (Zhipu)Apache-2.0 CogVideoX (~12.9k GitHub★) — loads natively in HF Diffusers, strong prompt following and smooth frames; a go-to for developers.
- 5LTX-Video (Lightricks)Lightricks' real-time DiT video model (~10.7k GitHub★) — notable for fast, production-capable generation and active releases.
Emerging
- 1HBAI-Ltd/Toonflow-appOpen-source desktop app that turns a novel or script into an animated short drama, chaining AI screenwriting, storyboarding and shot generation.
- 2vargHQ/sdkTypeScript SDK where you declare a video as JSX components, then it calls Kling, Sora or ElevenLabs and renders the result with FFmpeg.
- 3duixcom/Duix-AvatarOffline digital-human toolkit that trains an avatar from your own video and voice samples, then lip-syncs it to text on a local NVIDIA GPU.
- 4MeiGen-AI/InfiniteTalkOpen-weights research model that animates a photo or existing clip from an audio track, syncing lips, head and body for unlimited-length talking video.
- 5zhouxiaoka/autoclipSelf-hosted pipeline that transcribes a long video, lets an LLM pick highlights, and cuts them into short clips with titles and subtitles.
Skills
Skills you drop into a coding agent.Leading
- 1Claude Code Video Toolkit (digitalsamba)The standout Agent Skill for the space (~1.8k GitHub★): a skills bundle (Remotion, FFmpeg, voiceover, music) that scripts and renders an MP4 from one prompt.
- 2video-generation (ByteDance DeerFlow)Structured-JSON prompt → video skill from ByteDance's DeerFlow SuperAgent, with reference-image / first-last-frame guidance.
- 3claude-remotion-skill (haidrrrry)Open-source skill teaching Claude to create/edit motion-graphics videos with Remotion — B-roll, captions, sound design from a prompt. Thin/single-author.
Emerging
- 1Vincentwei1021/video-shotcraftAgent skill for Claude Code that animates product screenshots into 1080p promo videos with Remotion, from a catalogue of documented shot recipes.
- 2Clipcat-ai/clipcat-skill
- 3AgriciDaniel/lipnardoClaude Code skill wrapping the HeyGen API for avatar videos, adding batch runs, template variables, translation and cost estimates; needs a paid HeyGen key.
- 4thepandanlabs/video-gen-demo-booth
- 5karamusluk/reelsmith-claude-skill