Zhejiang University, Monash University, BAIR, and other institutions have released ATAS, a world-state-centric agentic video generation framework, paper number arXiv:2610.20571.
AtomicChat released an AD-Q4_K quantized version of Qwen-Image-2.1-Turbo, sized at 4.2GB — roughly 70% smaller than the official BF16 version at 14.2GB. The author measured LPIPS to be 21% lower than the standard Q4_K.
awesome-claude-video-skills collects 268 open-source projects that let coding agents make videos, divided into 10 categories by use such as frameworks and toolboxes, product promos, knowledge explainers, editing, and vertical short videos. Each project is marked with a SAFE or CAUTION safety rating, audited by Agent Skills Hub, and the list refreshes every 8 hours.
Alibaba's Tongyi open-sources Qwen-Image-2.1-Turbo, an accelerated checkpoint based on Qwen-Image-2.1 that uses the same 7B visual generation architecture; text-to-image and natural-language editing require only 8 denoising steps, and the official release says it can still generate 2K images and supports continuous editing such as adding accessories and changing scenes.
Why it matters: Qwen-Image-2.1-Turbo achieves 2K text-to-image and natural-language editing with 8 denoising steps, and offers two onboarding paths: Diffusers and API.
free-short-video v7.1.0 was released, adding staggered-retry guidance for AI video generation: when the upstream queue is full (HTTP 503 / queue-full), it suggests smoother time windows based on historical success rates, stored in UTC and displayed in each time zone's local time, with wording covering all 22 interface languages.
Open Film Skills v1.4.0 syncs the current source code and standalone download package, updates Storyboard Director to A5.7.2, and incorporates patches for scene front/reverse perspectives and prop structure details.
Because the official Qwen-Image 2.1 prompt rewriter is a 9B model requiring about 20 GB of memory, the author used ML Intern in HuggingChat to build a 0.8B version that runs on CPU, achieves a 99.7% valid output rate, and consumes about one quarter of the teacher model's tokens, with a total compute cost of $16 including 8,797 annotations.
The open-source project Director's Stage has released 0.4.11, adding physics capabilities including gravity, collision, inertia, ragdoll, and timed forces, along with physics demonstration scenes. Character appearance and basic actions now support selecting joints to adjust poses and recording keyframes, while also adding object grouping, alignment, arrays, and a user scene library that can be saved and reused.
AI Video Production Editor v3.0.0 released a macOS Apple Silicon DMG installer, which has been notarized by Apple with a notarization ticket included. No new installer is available for Windows yet; v2.7.0 EXE remains available.
free-short-video v7.0.6 has been released. Docker multi-architecture images (linux/amd64, linux/arm64) have been pushed to GHCR and Docker Hub, and can be obtained via docker pull ghcr.io/lcy362/free-short-video:7.0.6.
Teams from Tsinghua University and JD.com's Joy Future Academy have open-sourced Self Gradient Forcing Plus (SGF+), which resolves gradient conflicts in SGF by separating context writing and denoising parameters, enabling autoregressive video generation of up to 24 hours using only a 5-second training window.
pyvideotrans released a v4.15 Windows pre-packaged build; MacOS/Linux require source deployment. The new version updates the Deepgram.com channel models and adds speech synthesis support, updates the Elevenlabs.io models and removes models that will be deprecated, and adds subtitle dubbing silence buffer removal and intensity adjustment in the advanced options. It also adds the Audexum speech recognition channel, the Infistar translation channel, and Azerbaijani language support.
sirioberati released the Genjustsu open-source workflow, which replaces video subjects in a local UI using color depth, SAM 3 masks, and Enhancor's Seedance integration, producing a draft first and generating 1080p after approval.
Why it matters: The open-source local workflow replaces video subjects using color depth and SAM 3 masks while preserving the original audio track, so its reuse value can be evaluated.
A tweet roundup of 8 open-source GitHub projects that chain together the full pipeline from prompt to finished video: MoneyPrinterTurbo generates the script, assets, and music, and VoiceStudio handles local voice cloning and dubbing.
motionmaxxing is a skill that runs in Claude Code and any Agent that reads SKILL.md, reworking AI-generated promos, product ads, and kinetic typography pieces to motion design standards.
Mirage has added an open-source converter to its agent creative suite Tesseract, enabling Claude and ChatGPT to natively create Premiere (.PRPROJ) and After Effects (.AEP) files.
Toonflow 2.0.4 released, offering three installer downloads: Windows (64-bit), M-series Mac and Intel Mac. The M-series Mac installer has completed Developer ID signing and Apple notarization, though macOS may still prompt for confirmation on first open. See v2.0.3...v2.0.4 for the full changelog.
kandinskylab released the Kandinsky 6.0 Video series of diffusion models, including a 3B-parameter Lite and a 29B-parameter Pro, capable of generating 5-second videos with 44 kHz synchronized audio (including lip-sync), supporting text-to-audio-video and image-to-audio-video, and paired with a super-resolution model that upscales output to Full HD (1920×1080).
Why it matters: Kandinsky 6.0 has open-sourced video generation models with synchronized audio and lip-sync, and provides measured runtimes for various GPU tiers, making it easy for creators to assess local deployment costs.
Aurora Audio Studio 2.0.1 official version has been released for Windows and Mac; the Mac Apple Silicon installer has completed Developer ID signing and Apple notarization, while the Windows installer is unsigned.
unclejobs-ai has published motion-video-skill on GitHub, a Claude Code skill for creating 30- to 90-second animations and videos from scratch, breaking the process into six stages: story, characters, keyframes, video conversion, music and voiceover, and editing.
The open-source video generation project has released v4.0.6. Local reference images, first frames, and last frames now use original images, preserving dimensions, format, and alpha channels, without downscaling to 768 pixels or converting to JPEG. Docker has been upgraded to Node 22, excluding desktop artifacts and local environment configurations, and using system FFmpeg/FFprobe to fix the Linux ARM64 media probing unavailability issue.
The way to make videos with Claude Opus 5.5 is not for the model to directly generate frames, but to have Agent tools such as Claude Code and Codex write HTML, Canvas, WebGL, or Remotion animations, then export MP4 frame by frame, with subtitles, camera paths, and transitions all adjusted in code.
Why it matters: A video production path using Agents to write HTML, Canvas, and WebGL code and export MP4 frame by frame, with 10 finished pieces and 5 open-source projects available for reproduction.
facefusion released version v4-beta, adding a Timestamp alias type. This version was tagged by henryruhs on September 30, and specific feature details have not yet been elaborated in the release notes.
facefusion released version 3.9.1, updating to the latest onnxruntime version and limiting the arena workaround to affected onnxruntime versions. This version also fixes a performance issue with the content analyzer under CoreML.
DramaClip v0.2.0, the first public release, allows users to import a short drama with dozens of episodes and batch-generate ready-to-publish promotional short videos, with all media processing completed locally. This version uses OCR and ASR fusion transcription with hard-subtitle character-by-character correction—60% of segments were corrected in testing—and employs CAM++ voiceprint clustering for speaker diarization.
Why it matters: A short-drama promotional video tool that completes transcription, narration scripts, and voice cloning locally—explore its nine output modes and cut-point alignment approach.
free-short-video has released v7.0.5; images have been pushed to GHCR and Docker Hub, and can be installed via `docker pull ghcr.io/lcy362/free-short-video:7.0.5` or npm.
ReelMimic is an open-source project; users provide a reference video and a one-sentence request, and it breaks down editing rhythm, shot length, transitions, composition, color, and camera movement, first generating a storyboard and style-frame plan for confirmation, then handing it to up to 6 AI agents for parallel production, with each shot reviewed by another agent, and fixes requiring before-and-after comparison screenshots.
Why it matters: A locally run multi-agent animation tool that breaks down a reference video's rhythm and camera movement into a reviewable creative plan.
Second Unit released version 2026.09.28, updating the Windows download to the main branch and incorporating the workflow directories and fixes added after the September 15 package. The update includes MiniMax H3, LTX 2.5, SAM 3.1, WAN 2.2, and Juggernaut workflow directories, as well as model-specific prompt extensions, environment presets, seed handling, and an H3 Pose Follow fix.
VideoLingo v3.1.1 is released, adding a Fish Audio official dubbing provider that uses the s2.1-pro model by default and can clone a voice from 15–30 seconds of speech in the original video without saving the voice in the account.
VideoLingo v3.1.0 has been released; the default local speech recognition engine is changed to Qwen3-ASR, and Qwen3-ForcedAligner is used to align word timings, with the 1.7B model selected by default and the 0.6B model also switchable.
The open-source project killkli/minimax-h3-colab-skill is a standalone Codex skill that generates short MiniMax H3 Ref2VA videos from local reference images via Google Colab.
pyvideotrans releases v4.14 Windows pre-packaged version; the full package is 2.7G, and the 360MB patch package is only for overwriting the old package. This version optimizes speech, subtitle and image alignment algorithms, checks before Google Translate and automatically switches to Microsoft Translator when blocked by 429, and optimizes openai-whisper model downloads. MacOS/Linux require source deployment; if the new version has issues, users can continue using v4.13/v4.03.
The tensor shape mismatch when using Qwen-Image 2.1 together with the TaylorSeer cache has been fixed and merged into the Diffusers main branch, in under four days from the issue being filed on the evening of September 21 to the merge on September 25.
explainroo is a free open-source tool that lets coding agents generate explainer videos from a topic, with voiceover, timing and rendering all done on the local machine and the final output as an MP4. The agent only needs to write two files, script.md and scenes.js; Kokoro handles narration (28 voices, no account or API key needed), Whisper marks the pronunciation time of each word, Chrome draws the visuals in the background, and ffmpeg composites the video.
YeJe-cpu open-sourced SeeCut on GitHub, a skill for coding agents like Claude Code that automatically edits digital human or real-person voiceover footage into vertical short videos with motion effects, and can optionally push the result as a layered CapCut draft for fine-tuning.
Why it matters: The author open-sourced a reusable voiceover editing skill covering the full workflow from digital human voiceover to CapCut draft delivery.
bestagentkits open-sourced motion-video-skill on GitHub, a skill that lets an Agent generate beat-synced motion graphics videos, outputting at 1920×1080, 30 fps, with shot cuts landing on musical beats and reveal animations landing on narration words.
Tern is a podcast and video library search tool that runs locally on Apple Silicon, using three-way indexing via whisper.cpp (Whisper Large v3 Turbo, Q5), Apple Vision OCR, and SigLIP-2 to retrieve any clip by voice, on-screen text, or visual content, with results merged into a unified ranking.
BridgeMind released the open-source desktop app BridgeClip, which can find complete clips from podcasts, livestreams, and interviews, then edit and export subtitled vertical 9:16 or horizontal 16:9 short videos. Video rendering is done locally, AI processing uses the user's own OpenRouter key, and it supports macOS, Windows, and Linux, with the installer bundling Python, FFmpeg, and yt-dlp.
Why it matters: The open-source desktop tool chains long-video clipping, subtitling, and publishing into a single local rendering pipeline, enabling creators to evaluate how to repurpose their own assets.
SmartSub released v3.9.0, adding a built-in AI creation assistant (shortcut ⌘J / Ctrl+J) that supports context awareness, conversational subtitle polishing, screenshot troubleshooting, and file dialogue. This version supports the MCP protocol and CLI automation, providing 111 tools and a unified command line, with one-click connections to Cursor, Codex, and Claude Code, and a new audio-video toolbox covering audio extraction, subtitle merging/conversion/alignment, video compression, trimming, and GIF conversion.