August 5, 2026
Open Assets — a curated model catalog with one-click setup — plus link-based imports, dependable task output delivery, and a billing change for input downloads.
This release is about getting models into your workspace with less friction: a curated Open Assets catalog you can attach in one click, imports that link to the source instead of copying it, and a rework of task output delivery so results always make it back to you. It also changes how input download time is billed — see Policy changes.
Major features
Section titled “Major features”Open Assets — popular models, one click away
Section titled “Open Assets — popular models, one click away”Pod deploy and the Assets page now include Open assets: a curated catalog of widely used open models — FLUX, SDXL, Qwen, Wan, LTX-Video, and more — organized into ready-made bundles with per-template recommendations. Pick a set and it is registered in your workspace and attached to the right slots in one step; nothing to download locally or arrange by hand.
ComfyUI bundles also ship with matching starter workflows that appear inside ComfyUI ready to open, wired to the models the bundle installs. Read more →
You can also paste a Hugging Face or CivitAI URL straight into the deploy picker, and it becomes an attached asset in the same flow.
Link a model instead of copying it
Section titled “Link a model instead of copying it”Imports from Hugging Face and CivitAI can now register as a link rather than a copy. Registration completes immediately — no waiting for a transfer — and the files are downloaded straight into the Pod or task that uses them when it starts. Nothing is duplicated into your storage unless you choose to.
CivitAI API keys are now supported alongside Hugging Face tokens (add them under Settings › Integrations), gated repositories are flagged up front, and every link pins the exact upstream revision it was registered against — later changes at the source never silently alter your runs. Read more →
Task outputs, delivered dependably
Section titled “Task outputs, delivered dependably”Saving task outputs to Assets has been rebuilt: uploads now resume after interruptions and retry automatically for up to ~3 days, and the task page shows exactly where the transfer stands. A new task.outputs_ready webhook event fires once every file has landed — terminal events can arrive while outputs are still uploading, so this is the signal to fetch results programmatically. Don’t need the files? Discard outputs stops the transfer and releases the storage. Read more →
Your ComfyUI work saves itself
Section titled “Your ComfyUI work saves itself”Workflows you create or edit inside ComfyUI are harvested back into Assets automatically when the Pod stops — no manual export step. Bundled starter workflows are recognized and skipped unless you modified them, so the list stays your work. Read more →
Improvements
Section titled “Improvements”- Much faster model downloads. Large Hugging Face fetches now use a parallel transfer path — multi-gigabyte checkpoints that used to take most of an hour arrive in minutes.
- Live download progress, with resume. While a Pod fetches its image or input assets, the deploy card shows byte-level progress; interrupted fetches pick up where they left off and say so. Progress now also displays for digest-pinned images.
- A clearer input picker. Pod deploy shows destination paths first, groups files into
/inputssubdirectories, warns about filename collisions, and totals the download size before you start. - Scratch space managed for you. Pods deployed without a persistent volume get platform-managed scratch space for their declared paths that grows as it fills — ComfyUI-style templates work without pre-sizing a volume.
- Deploy cards grew up. Pods being created now show their price tag and live progress, and action buttons appear only once they can actually act.
- Hardware guidance at selection time. CPU and RAM choices warn inline when they fall outside the recommended band for the template.
- Stopping is immediate. Stopping a Pod during a long start or download now cancels within seconds and cleans up fully, instead of waiting out the operation.
Policy changes
Section titled “Policy changes”Input download time is billed as pod time
Section titled “Input download time is billed as pod time”Starting with this release, the time a Pod or task spends downloading input assets before your workload starts is billed at the same rate as running time — the hardware is reserved and your volumes are filling during that window. The deploy card and task page show the applicable rate during the download phase, and the billed window is capped (~6 hours).
- A download that fails from a platform fault is refunded — a task that never reached your code is refunded in full.
- A download the source rejects (an invalid or missing token, a gated repository you don’t have access to) is charged for the elapsed window.
- After a task ends, compute billing stops, but the
/outputsvolume continues to bill as storage until its files finish saving to Assets.
Terms of Service update
Section titled “Terms of Service update”The Terms now spell out how connected external services (model hubs and your own buckets) are handled: imports and publications run with your credentials, on your accounts, with Meshive acting as the transport. Compliance with each source’s terms and content licences remains yours.
- Image download progress no longer risks showing another Pod’s transfer when two Pods pull from the same repository at once, and stalled downloads are detected more reliably.
- Very long input downloads no longer fail at the last step after the transfer itself completed.
- A Pod whose creation failed at the final step could refuse to start afterwards — it now recovers cleanly.
- Deleting a Pod at the exact moment it finished creating no longer leaves a phantom card behind.
- Assets page columns no longer collapse at narrow window widths, and the import dialog no longer hides behind other panels.