A whole AI studio that refuses to leave your GPU

What it does OmniSpace is a full-stack, local-first AI creation workstation covering five modules: AI chat, comic generation, manga-drama (script-to-video) production, a novel writing desk, and knowledge learning with RAG. All inference runs on your own GPU by default — vLLM, llama.cpp (GGUF), and Transformers backends hot-swap behind a plugin system, with cloud APIs as an optional bring-your-own-key tier. The pitch is one machine, one tool, no data leaving the house.

The interesting bit The self-built VRAM coordinator (ledger, admission control, single orchestrator) claims to run the full creation pipeline on a 16GB consumer card by auto-loading, unloading, and yielding between models. Each heavy module also gets “bulkhead” isolation with a self-healing ladder — repeated failures trigger engine restarts, verification, and a safe-mode fallback.

Key highlights

• Full pipeline from script to storyboard to keyframes to video, with stale-asset badges when upstream edits invalidate old generations
• Comic features: four-view character sheets, five speech-bubble shapes, SAM-based background removal, PNG/PDF export
• Writing desk with chapter/word-count planning and three “de-AI-flavored” prose styles
• Serious ops layer: AES-GCM encrypted backups, power-loss recovery, audit logs, tiered delete confirmations, hardware health dashboard
• Claims 376 HTTP endpoints, 1132 pytest cases, and 101 vitest cases

Caveats

• Proprietary license, and the README states recent progress (except the plugin system) can’t be pushed to the repo due to licensing restrictions — so the public code may lag the described feature set
• Video generation leans on MiniMax H3 via local ComfyUI plus Wan2.2 — a heavy stack; the VRAM scheduler is described as “continuously tuned”
• Voice cloning is still in the research phase; comic-panel character LoRA consistency is listed as unfinished

Verdict Worth a look if you have a decent GPU and want an all-in-one local creation suite instead of stitching together ComfyUI, a chat UI, and a writing tool. If you wanted a permissively licensed codebase to build on, the proprietary license and repo lag are real friction.