Open-source IV Studio turns a 4090 PC into a full AI short-film studio on ComfyUI + MiniMax H3
itsVariance · reddit · 2026-11-03
Reddit user itsVariance released IV Studio (MIT), a Windows desktop app built on ComfyUI + MiniMax H3 that takes you from idea to finished short film on your own PC, entered in Comfy's Open Source Higgsfield challenge.
Key features:
- Prompt writing: a local Qwen3-VL (via Ollama) or any MCP-connected agent inspects reference images and writes H3 prompts; a checker auto-fixes common H3 failure modes (frame-grid lengths, speaker tags, unattached references)
- Hunt then finish: quick previews ranked by Whisper-heard dialogue, likeness and sharpness; picks finish from the same latent with DLSS 5 or SeedVR2
- Projects: a six-step wizard from script and casting to storyboard sketches playable as animatics; every shot is generated from its sketch for consistent faces and rooms
- Staging board: place actors and cameras on a top-down floor plan, converted to screen directions or a grey previs clip H3 follows
- Editing: right-click to extend shots, fill gaps, retake lines, cut people out (SAM 3.1), reframe to vertical
- Also: character LoRA training, MiniMax Music 3 / ACE-Step music, VoxCPM2 local voices, Civitai LoRAs, and a 142-tool MCP server so Claude Code can co-direct
No strong GPU? It can render on your own Runpod or Modal account, powering the GPU off when idle. Tested on a 4090 with 64 GB RAM; 16 GB cards work with smaller previews.
More from coding & agent
- First hands-on: Meta's Muse Wonder 1.0 vs DeepSeek V4.1 Flash in Codex CLI — Richidm · 2026-11-03
- "Claude Code has been terrible today" — developer smells a new Opus launch coming — hermanschutte · 2026-11-03
- One command audits every skill loaded by Claude Code, Codex, Cursor and more — 236 found — UnicornAdoption · 2026-11-03
- When Agent Swarms Rewrite Your Backend in a Language You Don't Know — smlpth · 2026-11-03
- oxc parser gets up to 20% faster in next release, cutting rolldown bundling time ~6% — irvinebroque · 2026-11-03
- Qwen's Astra gains come from an agentic ML loop, not weight introspection — petrusenko_max · 2026-11-03