September 4, 2026
Quick Deploy presets now start from the standard ComfyUI image and attach their model set on deploy — no storage step — plus sturdier large-image Pods, right-sized scratch disks, and allowance emails.
Major features
Section titled “Major features”Quick Deploy: model sets instead of pre-baked images
Section titled “Quick Deploy: model sets instead of pre-baked images”Every Quick Deploy preset — FLUX, Qwen-Image, Z-Image Turbo, WAN 2.2 and LTX-Video — now launches the standard ComfyUI image and attaches the preset’s official model set automatically when the Pod is created. The model files are the same bundles you can browse under Assets, so what you get is always the current, curated set rather than whatever was baked into an image months ago.
What changes on the Review & Deploy page:
- The storage step is gone. The model files land on working storage that is provisioned for the Pod automatically; its cost appears as its own line in the total, so the Deploy button and the receipt always agree.
- A Model Set card lists exactly which files will be attached and their total size. Download time is billed at the Pod rate, and if a download fails the Pod is rolled back.
- Presets that ship a starter workflow show it on the card; the workflow appears in ComfyUI’s sidebar once the Pod is running, so the first generation is a click away.
Pods you already created from the older pre-baked presets keep running unchanged. Quick Deploy →
Improvements
Section titled “Improvements”- Scratch disks grow by what is needed. Working storage that the platform provisions for a Pod used to grow in fixed steps of about 10 GB, so a small asset could reserve — and bill — far more than it used. Growth is now sized to the shortfall plus a modest margin. Thresholds are tuned and may change.
- Storage allowance emails. When a Pod’s scratch disk reaches the storage allowance assigned to it and can no longer grow, workspace admins and billing members receive an email naming the Pod and path, so you can free space or attach a volume before writes start failing.
- The meshive.ai landing page gained an About page and refreshed site metadata.
- Pods with very large images start reliably. A Pod whose image took a long time to download — tens of gigabytes — could be torn down by the platform right after it started, because the long wait left the Pod’s bookkeeping stale. Image pulls of any length now complete cleanly.
- Serving deployments that spend a long time downloading model weights were exposed to the same issue and are fixed the same way.