Pods you can start from
Each one is a GPU box we have already described: a workload, whatever that box has installed on first boot, and a set of model weights downloaded onto its disk before you open it. Everything you add afterwards stays on that disk, so stopping the box does not mean building it again.
Plain boxes
Image generation
ComfyUI with SDXL ControlNet
A GPU box with ComfyUI already installed and the OpenPose ControlNet weights already pulled down.
ComfyUI preinstalledComfyUI with NoobAI-XL
A GPU box with ComfyUI installed and the NoobAI-XL v-prediction checkpoint already pulled down.
ComfyUI preinstalledComfyUI for Wan 2.2 video
A large-memory box with ComfyUI installed and the Wan 2.2 TI2V weights already on disk.
ComfyUI preinstalledNotebooks
A notebook with Qwen3-8B on disk
JupyterLab and PyTorch already installed, with the Qwen3-8B weights already pulled down.
JupyterLab + PyTorch preinstalledA notebook with Qwen3-VL on disk
A JupyterLab box holding the Qwen3-VL-4B vision-language weights, already downloaded.
JupyterLab + PyTorch preinstalledGuides for what you run on a pod
Serving a model from a pod? Start with serving LLMs with vLLM or choosing between vLLM, TensorRT-LLM and SGLang. Running image workflows? See the ComfyUI guides. Keeping a pod between sessions is covered under Stop & Start and Volumes.
Start an empty box insteadThese 6 pods are ours. The console can already hand a saved box to someone else as a private link, and publishing those here is where this page goes next. A shelf of other people's pods needs pods on it first, so this one starts with ours.
Models to run on an image-generation pod: GPU requirements for Stable Diffusion XL, GPU requirements for Wan2.1 VACE 14B, GPU requirements for RealVisXL V5.0, and Best open image generation models. Browse the full lists: VRAM calculator for every GPU and GPU prices and specs for every card.
