Three open-source video tools picking up steam: Mago Studio, Rigby2D and a MiniMax H3 'short-drama workbench' for ComfyUI

Mago Studio (AGPL-3.0, about 420 stars nine days after the repo was created). The local-first desktop version of the Mago AI video editor, for Windows, macOS and Linux. 'Director' goes from script to cast, shot list, storyboards and scenes where each shot continues from the last. 'Transform' restyles footage frame-accurately: frame ranges, masked partial edits, relighting and character replacement. It also has versioning and review tools. On Windows a setup wizard installs a portable ComfyUI with MiniMax H3 (4-step turbo and 8-step quality LoRAs), H3 VFX LoRAs and Qwen3-VL 4B for captions, and needs an NVIDIA GPU with 8 GB of VRAM and about 63 GB of disk, per the README. On Mac and Linux you connect your own ComfyUI. Cloud models are optional and paid. Some of the bundled LoRAs swap heads and characters, so use them only on footage and people you have the rights and consent for, and note that the MiniMax H3 licence applies on top of the app's.

Rigby2D (Apache-2.0, about 220 stars, v0.5.1 shipped Nov 11). A free alternative to Spine, DragonBones and Live2D-style rigs. AutoRig turns a layered PSD or SVG into bones, bending meshes and IK in one step. Its Film workspace has 4K output, an animatable camera, audio tracks and viseme lip-sync from on-device Whisper transcription, and renders to PNG sequences, WebM with alpha or MP4. A built-in MCP server (a .mcpb bundle for Claude Desktop) lets an AI assistant inspect, rig and animate your project. It runs in the browser or as a desktop app. For explainer and kids'-content teams, it's an open way to keep a character on-model across episodes without generating every frame.

comfyui-AICG3D, the 'short-drama workbench' (GPL-3.0, about 515 stars, updated Nov 11). A Chinese-language ComfyUI pack built around MiniMax H3 for local short dramas and long videos made in segments. The latest release (v1.1.27) moves the second-pass upscale and quality fix to after all segments are joined, which avoids seams, stutter and exposure jumps between parts. It also adds exposure locking and seam stabilisation, and keeps each character's cloned voice consistent from segment to segment. Workflows are in the repo. The same author also publishes a pose-and-outfit node pack (comfyui-aicg3d-PoseSwap, picked by @GitHub_Daily today); because it includes face swapping from a third-party image library, we're only noting it, and the author also says not to use it commercially or to pass anyone off as a real person.

Related tip from the official @ComfyUI account (Nov 11): a short breakdown of what Seedance 2.5 and MiniMax H3 actually take from a 3D blockout, from plain shapes up to a fully animated character. It's worth watching if you use blockout-to-video workflows like Comfy Previs or the Blender-to-Seedance template already in our directory.

Signals (not hot by our rule). Mago Studio: GitHub stars plus scattered posts on X (e.g. TheSoloFactory). Rigby2D: active releases and community posts. comfyui-AICG3D: Chinese-community traction and frequent releases. None is on GitHub Trending today or posted by more than one tracked source. Feature claims come from each project's README. We haven't run them.

Our take: Mago Studio is the one to watch if you want a Runway-style editing workflow on your own GPU. Rigby2D is the practical pick for recurring 2D characters with lip-sync, and AICG3D is for teams already deep in MiniMax H3 who need minute-plus pieces that don't fall apart at the joins.