Quick Deploy - ComfyUI

Quick Deploy is the fastest way to get a ComfyUI workspace running on Meshive. You pick a model set, Meshive auto-recommends the cheapest GPU that fits it, and you click Deploy — the set is downloaded into the pod for you, and a starter workflow is waiting in ComfyUI’s sidebar.

Normally, running a model in ComfyUI means SSH-ing into the pod, downloading the right .safetensors files, dropping each into the correct models/ subfolder, and only then loading it in a workflow. Quick Deploy removes that entire step: every card pairs the standard ComfyUI image with an official model set from Assets, placed in the right folders at deploy time — no downloads to manage, no folder wrangling.

Quick Deploy vs. the Pod wizard

Use Quick Deploy when you want a ready-to-run model set with sensible defaults. Use the full Pod → Deploy wizard (see Deploying ComfyUI) when you need to customize the image tag, command, environment variables, endpoints, attach several sets at once, or bring your own weights.

Available model sets

Every card on the Quick Deploy page is one official model set from the Asset Hub catalog — the same sets you can attach by hand in the full wizard. Publishing a set adds a card, so the list grows without a release.

Cards can be filtered by what the model produces — Image (FLUX.1, SDXL, Qwen-Image, Qwen-Image Edit, Z-Image Turbo) and Video (Wan 2.1 / 2.2, LTX-Video, MiniMax H3) — using the tabs at the top of the list. The list shows all sets by default.

Each card carries the four things that decide whether a set will work for you:

On the card	What it means
Image / Video	What the model produces
N GB VRAM	The minimum the set needs. GPUs below this are not offered — there is no under-spec fallback.
Resolution	The resolution that figure is based on (the set’s sweet spot). Shown when the set declares one.
Download size	How much is fetched before ComfyUI starts. Sets over about 25 GB are marked Large (this threshold is tuned and may change).

Note

Weights ship FP8-quantized where the publisher offers it, so VRAM is roughly half of FP16. The exact set of cards depends on which sets are currently published, so your console may show more or fewer than the families listed above. Model sets are curated and may be updated between releases — the Model Set section of the deploy panel always shows what will be attached today.

Deploy

1. Open Quick Deploy

   In your Workspace sidebar, click Quick Deploy under Resources. This is a single screen: the model set cards on the left, and a deploy panel on the right that already shows a recommended set (and its recommended GPU) before you click anything. Use the All / Image / Video tabs above the list to filter which cards are shown.

2. Pick a model set and confirm

   Click the card you want (for example FLUX.1-schnell) — the panel on the right updates in place to that set. It’s pre-filled with a recommended configuration; you only change what you want:

   • Pod Name — pre-filled with a suggested name; edit if you like.
   • Rental Type — Demand (uninterrupted) or Spot (cheaper, may be reallocated).
   • GPU — auto-selected as the cheapest GPU that meets the set’s VRAM requirement, with 1 GPU. Pick a different card from the GPU list to switch — each option shows the GPU model, VRAM, vCPU/RAM, and the hourly price.
   • Model Set — the official bundle that will be attached: its name, the number of files and total size, and a Starter workflow included badge when the set ships one. The files are downloaded to the pod on deploy; download time is billed at the pod rate, and if the download fails the pod is rolled back.
   • Specs — VRAM, vCPU, RAM, and System Storage are derived automatically from the chosen GPU.
   • Total — the hourly price updates live as you change options. It includes the working storage provisioned for the model set, priced by allocated size like local storage.

   Click Deploy.

   No storage step

   There is nothing to attach. The model files land on working storage that Meshive provisions for the pod automatically and grows as you fill it — see Storage. To keep custom models, LoRAs, or generated outputs across pods, deploy the base ComfyUI template from the full wizard and attach a volume there.

3. Wait for the pod to start

   After you click Deploy, Meshive schedules the pod, pulls the ComfyUI image, and downloads the model set — the status moves through Creating → Pulling container image → Running, with download progress shown along the way. Model sets run from about 6 GB to 40 GB, so allow a few minutes on the first deploy — longer for a set marked Large.

4. Connect to ComfyUI

   Once the pod shows Running, open the Pod page and click the Connect (plug) icon to launch the ComfyUI GUI.

5. Open the starter workflow

   Sets that include a starter workflow show it in ComfyUI’s sidebar — open it, type your prompt, and queue. Because the model files are already in place, the graph runs without the “missing models” download prompt.

   You can also use ComfyUI’s own template browser — Workflow → Browse Templates — and pick any template for that model. See Bundled ComfyUI workflows for how bundled workflows behave.

Tip

ComfyUI also exposes a Jupyter endpoint (port 8888) and the ComfyUI API server (port 8080) alongside the GUI (port 3000), so you can drop into a terminal or drive it programmatically.

Next

• Load a workflow and generate your first output — for the base ComfyUI template, this also covers downloading model weights yourself.