October 18, 2026
MOSS-Image joins Quick Deploy with text-to-image and layout-to-image workflows, volumes stay available until a Pod has fully shut down, and asset injection into a running Pod recovers from temporary errors.
Major features
Section titled “Major features”MOSS-Image in Quick Deploy
Section titled “MOSS-Image in Quick Deploy”MOSS-Image from OpenMOSS is now a Quick Deploy card, with two official ComfyUI workflows in the sidebar:
- Text to Image — an optional prompt upsampler expands a short prompt into one detailed caption before it reaches the model. Turn it off to send your prompt as written.
- Layout to Image — draw boxes on the canvas and describe what goes in each one, including words to render as text, to control where things appear.
The set is about 15.5 GB, needs a GPU with at least 14 GB of memory, and downloads without a Hugging Face token. It sits under New at the top of the Quick Deploy list for the next few weeks.
License: Apache-2.0.
- Volumes stay available while a Pod shuts down. When you delete a Pod, its volumes now stay attached until the Pod has fully exited. Apps that save a final checkpoint or flush files after the shutdown signal no longer fail with
Stale file handleon network volumes, and on encrypted local volumes those last writes are no longer lost without an error. Deleting a Pod can take a little longer as a result, usually a few seconds. - Stopping or updating a Pod no longer cuts off a shared network volume. Stopping, restarting or updating a Pod used to detach a network volume on that machine before the Pod was down, which could also interrupt another Pod on the same machine using the same volume. The volume is now released only after the Pod has exited.
- Asset injection recovers from temporary errors. If a temporary server error interrupts attaching an asset to a running Pod, the injection now retries instead of being marked as failed for good.
For hosts
Section titled “For hosts”- Safer volume clean-up in the node agent. When a Pod is stuck shutting down, the agent now detaches a network volume only for that Pod, and never deletes data under a path that is still mounted. The new agent installed itself on online compute machines today with a short agent restart; running Pods were not affected.