Skip to content
ChatGPT & LLMs

Install Modly v0.4.2 for Local AI 3D Modeling

Deploy Modly Beta v0.4.2 for offline AI 3D modeling: system specs, installers, first extension, verify steps, and real install fixes.

6 min readIntermediate

Cloud AI 3D tabs nick you per mesh. Reference photos leave your machine. AI 3D modeling on a GPU you own stops both. This guide puts Modly Beta v0.4.2 (GitHub release 28 Aug 2026) on disk – open-source desktop app, image-to-mesh offline once weights are local, no credit meter.

Installers, first extension, a real health check, and the GitHub failures people actually open issues for.

System requirements before you download

Hardware gates which extensions run. The shell can open on a weak box and still refuse generation.

Item Minimum Recommended
OS Windows, modern Linux, or macOS on Apple Silicon Same; current GPU drivers
GPU NVIDIA CUDA GPU; ~6 GB+ VRAM for Hunyuan3D 2 Mini (mesh-oriented) 8-12 GB+ VRAM for heavier official extensions; GGUF variants cut VRAM pressure
macOS Apple Silicon only (arm64 build) More unified memory helps larger backends
Disk Room for the app plus one weight pack (several GB) Extra free space if you keep multiple packs

Generation path is NVIDIA CUDA per the official FAQ (as of the v0.4.2 docs set). AMD is not a supported route. Intel Macs are out – shipping macOS asset is arm64 only.

Funny thing about “unlimited local meshes”: you trade the SaaS invoice for thermals, driver drama, and a multi-GB download the first time. If that trade still feels fair, keep going.

Official download source (v0.4.2)

Binaries from GitHub Releases or the project site only.

Assets labeled for v0.4.2 on the expanded release page:

  • Modly-Setup-0.4.2.exe (~180 MB) – Windows
  • Modly-0.4.2.AppImage (~205 MB) – Linux
  • Modly-0.4.2-arm64.dmg (~191 MB) – Apple Silicon macOS

Changelog worth caring about: Linux/Mac launch script fix. UTF-8 Windows stdio (UnicodeEncodeError quieted). Extension install lifecycle fixes. Shared cuda→mps→cpu device helper. All from the v0.4.2 release notes.

Install Modly step by step

Installer first. Source is for people changing the code.

Windows (recommended path)

  1. Get Modly-Setup-0.4.2.exe from the latest release.
  2. Run setup; finish.
  3. Start Modly. First boot may unpack bundled Python – let it finish.
  4. AV sometimes quarantines embedded Python and you only notice when venv creation dies. Exclusion on the Modly install folder; retry. Matches the pattern around issue #255 / related install threads in community reports.

Linux (AppImage)

chmod +x Modly-0.4.2.AppImage
./Modly-0.4.2.AppImage

Minimal distros may need basic GL libraries. Keep the NVIDIA driver stack current.

Apple Silicon macOS

  1. Open Modly-0.4.2-arm64.dmg.
  2. Drag Modly into Applications.
  3. If macOS blocks first launch, allow it under Privacy & Security, then reopen.

Alternative: run from source

git clone https://github.com/lightningpixel/modly.git
cd modly
npm install
cd api
python -m venv .venv
# Windows: .venvScriptsactivate
source .venv/bin/activate # Linux/macOS
pip install -r requirements.txt
cd ..
npm run dev

Windows source/clone is where python-embed + missing venv module pain shows up. The .exe path dodges most of that. Use source when you need to patch code – not for a first mesh.

Pro tip: One light official extension first – Hunyuan3D 2 Mini. Full Image → mesh once. Then stack Turbo / TripoSG / Trellis2 GGUF. Multi-extension chaos is usually a bad first venv, not a dead GPU.

First-time configuration: extensions + minimal workflow

Bare install = no generator. Extensions are GitHub repos you add from the UI.

  1. Models → Install from GitHub.
  2. Paste an official HTTPS URL, e.g. https://github.com/lightningpixel/modly-hunyuan3d-mini-extension.
  3. Confirm; wait for the extension venv build.
  4. Download the weights/variant the UI offers (plan for several GB).
  5. Workflows: Image → Generate Mesh → Add to Scene (README minimal workflow). Bad links → toasts, not a quiet empty viewport.
  6. Generate: pick that workflow, drop a clean still (subject centered, boring background), run Generate 3D Model.
  7. If nothing shows, open logs/errors in the app before reinstalling everything.

Same install flow for other official extensions listed in the project README (Mini Turbo/Fast, TripoSG, Trellis2 GGUF, etc.). Weights on disk ⇒ offline runs. No upload path for generation after that.

Verify the install works

Don’t guess. Run these three.

  1. UI: app opens; Models lists the extension; weights show present.
  2. End-to-end: one photo → preview mesh → export GLB (or STL if you care about print tests). Exports called out on the site/FAQ: GLB, OBJ, STL, PLY.
  3. CLI (app running; checkout that still has the tools tree):
python tools/modly-cli/agent.py health

Health should print machine-readable JSON on stdout – README Modly CLI section.

When that first GLB looks like a potato with heroic topology… is the install “done,” or are you just at the start of mesh cleanup? Both, honestly. The app did its job; Blender or your slicer does the next one.

Common install errors and fixes

“No module named venv” / python-embed (Windows, source or early backend setup). Embeddable Python is not a full venv host. Use Modly-Setup-0.4.2.exe. If logs show AV locking python-embedpython.exe, exclude that path and retry – same class of report as issue #255.

AppImage: works once, breaks after restart (backend python path gone). Extension venvs can symlink into the AppImage’s ephemeral mount; that mount path changes next launch. Recreate/Repair the extension from Models, or treat symlinked venvs as fragile (issue #210). Some setups discuss --copies-style venvs so the env does not point into the disposable mount.

CUDA OOM mid-run. Drop to Hunyuan3D Mini or a GGUF/quantized Trellis-style variant, close other GPU apps, lower resolution if the extension exposes it. FAQ line: smaller or quantized variant first.

RTX 50-series (Blackwell) kernel failures on some Trellis2/GGUF paths. Community/extension runtime messages point at too-old toolchain pieces for newer SM targets – install a current CUDA toolkit so a modern ptxas is on PATH, restart Modly, Repair the extension. Treat exact toolkit minors as “check the extension’s own error text,” because they move.

Extension install dies immediately. HTTPS URL to a repo the installer accepts; disk full kills weight pulls halfway. Free space before you blame the GPU.

Upgrade and uninstall

Upgrade: newer installer from Releases over the old one (Windows setup / replace AppImage / new dmg). Open Models; Repair extensions still glued to stale venvs. Git checkouts: git pull, npm install, refresh api deps with pip install -r requirements.txt.

No special 0.4.x → 0.4.2 migration file beyond extension repair. Weights usually stay.

Uninstall:

  • Windows: Settings → Apps → Modly → Uninstall. Delete leftover extension/weight folders under the user Modly data dir if you want the GBs gone.
  • Linux: remove the AppImage; clear config/extension dirs it created under your home.
  • macOS: trash the app; clear Application Support data for a clean slate.

Source: delete the clone, node_modules, and api/.venv.

Next move: grab your platform asset for v0.4.2, install Hunyuan3D Mini from GitHub inside Models, one photo through Image → Generate Mesh → Add to Scene. Touch other extensions after that succeeds.

FAQ

Is Modly free after install?

Yes. MIT app, no credits, no subscription. You pay GPU time and disk for weights.

Can I ship commercial assets from Modly meshes?

App license is MIT. Model extensions are not. Hunyuan community terms are the usual example – fine for tinkering, not a blanket client-delivery green light. Before you invoice anyone, read that extension’s README and assume commercial clearance is per-model. Installing Modly does not wash the weights.

Why does my AppImage need Repair every reboot?

Ephemeral AppImage mounts + extension venvs. See the AppImage fix under Common install errors (issue #210) – Repair against the current mount, or avoid symlink-into-mount venvs.

Related AI Tools