Asset Hub

One platform for every AI-generated asset.

The central hub where AI generators and LLMs create, refine, and manage your digital assets — images, video, audio, 3D and splats — behind one API and one interface, across cloud and on-premise.

View on GitHub Learn more

One home for AI asset creation.
No more juggling a dozen tools.

AI generation is fragmented — a different tool, account, and API for every image model, 3D service, and LLM. Asset Hub brings them together: a single platform that orchestrates every generator behind one API, stores every result in one searchable library, and enriches it automatically with LLM-generated metadata. Create, refine, organize, and ship — without ever leaving the hub.


A 30-second tour.

From a single prompt to a fully-tagged, searchable library — library, generation, activity, search and settings, all in one local-first workspace.

Recorded on a local instance · ComfyUI + Ollama, no cloud required.

View the full walkthrough

One UI. Every format.

From a text prompt or a source image to a finished, production-ready asset — in the format you need.

Images

Generation, editing & textures

Video

Text-, image- & video-to-video

Audio

Speech, music & SFX

3D Models

Meshing, retexture & remesh

Splats

Gaussian capture & NeRF


Everything between the prompt and the asset.

Unified API

One REST endpoint orchestrates every backend. Send a request, get an asset — the hub handles routing, polling, and delivery.

Multi-provider engine

Meshy, Tripo3D, ComfyUI, Ollama and OpenAI plug in through one provider registry — cloud or fully local, side by side.

Generation pipeline

Background workers track every job from pending to preview to final — auto-downloading results and refining meshes as they land.

Semantic search

Hybrid vector + keyword search across your whole library. Find any asset by what it looks like, not just what it's called.

Collections & bulk jobs

Group assets into sets and run a whole collection through generation at once — one click turns a folder of images into 3D.

LLM metadata enrichment

Every finished asset is auto-described, tagged, and categorized by an LLM — so your library stays organized without manual work.

Auth built in

JWT sessions, personal API keys, and optional OIDC / SSO — with admin approval flows and short-lived signed asset URLs.

Self-hostable

Ships as Docker Compose — FastAPI, worker, PostgreSQL and web UI. Run it entirely on your own hardware, your data stays yours.

One library, every asset

Images, video, audio, meshes and splats live together — versioned, previewable, and downloadable from a single dashboard.


Security is not an afterthought.

Self-hosted means your prompts, assets and keys stay on your own infrastructure. Asset Hub is hardened by default and was reviewed end to end — with an automated security test suite guarding every guarantee.

Encrypted secrets

Provider API keys are encrypted at rest, and the app refuses to start on weak or default secrets.

SSRF-guarded

Every outbound fetch to a provider or reference image is validated; access to private networks is blocked.

Per-user isolation

Ownership is enforced on every asset, set and job — users only ever see their own work.

Revocable sessions

Refresh tokens rotate on every use; a password change instantly revokes all of the account's sessions.

Proxy-isolated sharing

Public share links run through a dedicated, credential-less proxy — expiring, password-protectable, revocable and rate-limited.

Hardened runtime

Non-root containers with dropped capabilities, security headers, an SSO option and auth rate limiting.


Bring your own generators.

Cloud services and local models, orchestrated together through one provider registry.

Meshy AI Tripo3D ComfyUI Ollama OpenAI & more

The roadmap.

A focused path from a hardened multi-format core, through a polished 1.0, to a self-extending creation platform.

Shipped
  • Multi-format generationImages, text-to-speech & 3D from one UI
  • Multi-provider "Race" modeSame task, several providers, compared side by side
  • Semantic library searchHybrid vector + keyword, paginated
  • Collections & bulk generationRun whole sets through generation at once
  • Pluggable provider engineAdd cloud or local providers via templates
  • Hardened & self-hostableSecurity-reviewed, SSO, one-command Docker
  • Secure public share linksProxy-isolated, expiring, password-protected & revocable
Road to 1.0
  • Live activity feedScans, uploads & downloads in real time
  • “All → splat” bundlesTurn a whole collection into one splat capture
  • API hardeningPer-route rate limits & idempotency keys
  • More provider templatesA1111 & a generic REST integration
  • One-command deploysAutomated updates & a clean upgrade path
  • Public hosted demoTry Asset Hub without installing anything
Beyond
  • Intelligent image-to-imageA VLM reads your intent and picks the edit
  • Workflow chainingOne input flows through several models automatically
  • Self-integrating providersThe LLM inspects a new API or workflow and wires it up itself
  • Smart collection rulesAdmission & auto-conversion on the assets you add
  • Scale-outMulti-instance rate limiting & cursor pagination

Start building your asset hub.

Open source and self-hostable. Clone it, run docker compose up, and start generating.