generative-video

AI filmmaking on an infinite node canvas. Generate, build characters, and train LoRAs on your own GPU.

Run MiniMax Hailuo H3 and LTX-2.5 video generation locally on a Mac. Joint audio+video, character LoRA training, one-click Pinokio install. MLX — no CUDA, no cloud, no API key.

Hermes Agent skill for typography-led lyric videos using GPT Image 2, Gemini Omni Flash, and exact-text QC

Open-source local implementation of MiniMax H3 Context-IR. Turn simple prompts into structured, validated H3 video briefs.

Turn a story into a finished film — on your own Mac, no cloud. Story Forge is a local-only generative cinema pipeline: stills, motion, voices, music, grade, and sound, start to finish. Flux · LTX-2 · Piper · ACE-Step · ffmpeg — built for and running 100% on a Mac (M5, 128GB).

Beyond Time Shifts (ECCV 2026): a reference-free audio-visual synchronization evaluator built on Qwen2.5-Omni-3B, with SynthSync data, R-GRPO training, and the SyncBench benchmark.

ComfyUI custom-node pack + empirical prompt-sensitivity study for LTX 2.3 video generation. Three nodes (CameraBlock, AudioLine, PromptComposer) + 3 findings + ready-to-run example workflows.

Generate high-quality videos from text prompts and optional images using the Wan 2.2 TI2V-5B model.

Agent Skill that turns prompts into 9:16 animated explainer shorts — entirely in code (Python → SVG → ffmpeg, no browser, no After Effects).

Generate high-quality videos from first and last frame images combined with text prompts using the Wan2.1-FLF2V-14B-720P model.

AI-powered finger-frame video transformation with stabilized hand tracking, perspective portals, reference-guided video restyling, occlusion, and cinematic portal transitions.

Production control skills and deterministic quality gates for AI filmmaking. External pilots wanted.

Local-first video diagnostics and guided repair with frame-level evidence, safe sharing, useful-content extraction, and optional AI.

Orithet is a cutting-edge generative video art tool that seamlessly blends four conceptual systems to create mesmerizing, algorithmic visual experiences:

Claude Code skill: phone-realistic UGC reaction ads (Higgsfield generation + Remotion), native-audio-first.

A generative film studio modeled as a real production studio in layers (roles -> grounding -> code), orchestrating multiple AI backends per department (video on Gemini Omni, stills on Azure gpt-image) - grounded in film craft, starting with Bowen's Grammar of the Shot.

Open benchmark for generative video models, judged on craft by working filmmakers and engineers.

Comparative evaluation of six video reskin/editing methods on one production shot -- research evaluation, not a production pipeline

Stateless MCP server that keeps AI-generated characters, locations, and shots consistent across a film/video project — where the production stands, per-shot status, and marking shots fired/locked. For Higgsfield, Runway, Kling, Seedance, nano-banana, ElevenLabs.

Advanced 6DoF camera paths, visualization, chaining, and multi-frame reference conditioning for native ComfyUI Wan workflows