☁️BizyAir LayerMask: Load SAM2 Model(Advance)

If you've used chflame163's ComfyUI_LayerStyle pack (the one everyone just calls "dzNodes" after its category prefix), you've probably run into "LayerMask: Load SAM2 Model" already - it's the loader that feeds Meta's SAM2 into the LayerMask family of masking nodes for things like point-prompted object cutouts and person segmentation. This BizyAir node is the cloud version of exactly that. Same job, same output type, except the actual SAM2 weights live on SiliconFlow's servers instead of your models/sams folder.

That's the whole pitch for this node, and honestly for most of BizyAir's catalog: it takes an existing, often GPU- and disk-hungry loader and turns it into a thin cloud proxy. SAM2 checkpoints aren't huge by video-model standards, but if you're already tight on VRAM running a big diffusion model, offloading segmentation to the cloud frees that memory for the part that actually needs it.

What it does

This node loads a Segment Anything 2 model and hands you back a model handle that other LayerMask nodes (segment-by-point, segment-by-text, whatever downstream masking node you're chaining it into) will consume. SAM2 itself, if you're not familiar, is Meta's follow-up to the original Segment Anything - same "click a point or draw a box, get a precise mask" workflow, but SAM2 added temporal consistency so it can track and segment the same object across video frames, not just single images.

Inputs and outputs

There are exactly three required inputs, and none of them are things you'll spend much time tweaking:

• sam2_model - which SAM2 checkpoint to load. As of this writing there's just the one option, sam2.1_hiera_base_plus.safetensors, so this dropdown isn't much of a decision.
• precision - bf16. Again, one option - BizyAir picked the precision for you.
• device - cuda. Since execution happens on BizyAir's GPUs regardless of what's in your machine, this is mostly vestigial.

The single output is sam2_model, typed LS_SAM2_MODEL. That type prefix matters: it's the same custom type chflame163's LayerStyle pack uses, which means this output is built to plug into other LayerMask-family nodes expecting LS_SAM2_MODEL specifically - not a generic SAM_MODEL slot. If you're mixing BizyAir nodes with local LayerStyle nodes in the same graph, double-check the type actually matches before you go hunting for why the connection won't snap in.

How to install it

Grab the whole pack, not just this node - BizyAir ships as one custom_nodes install and this loader comes along with everything else.

Via ComfyUI Manager: open the Manager, search "BizyAir," hit Install, restart.

Via git clone:

cd ComfyUI/custom_nodes
git clone https://github.com/siliconflow/BizyAir.git

Restart ComfyUI after either method.

Before you generate anything: BizyAir nodes need an account and API key set once - the README points you at their login flow to grab one. Skip this step and every BizyAir node in your graph, this one included, will just fail at run time instead of loading.

Common issues

The single-choice dropdowns here are a hint about how BizyAir works generally: because everything actually runs on their infrastructure, the pack only exposes the checkpoint and precision combos SiliconFlow has actually deployed server-side. You can't point this node at a SAM2 checkpoint you trained yourself or one that isn't on their list - that's the trade-off for not managing the model file locally.

If you've never touched BizyAir before and this node showed up in your workflow unannounced, it's worth knowing the pack has occasionally gotten auto-installed by ComfyUI Manager when it mis-guessed which pack a missing node belonged to (a registry issue that's since been fixed). It's not malware - it's a legitimate, if very China-cloud-flavored, node pack - but if you don't actually want cloud execution, it's safe to uninstall via the Manager and delete the folder.

Beyond that, the failure mode to expect is API-key related: no key set, expired key, or insufficient BizyAir credits will surface as a run-time error on this node (or whatever runs after it) rather than anything you can debug locally. Check your BizyAir account status before assuming your graph is wired wrong.