☁️BizyAir Upscale Image (using Model)

This is the "apply" half of a two-node pair - it's not the node where you pick an upscale model, it's the node that actually runs one. You'll pair it with an upscale model loader (also in this pack, not covered here) that hands you a BIZYAIR_UPSCALE_MODEL, then plug that plus your image into this node to get the upscaled result.

What kind of upscaling this is

Worth being precise here, because "upscale model" covers a wide range of tools these days. This node's input type is BIZYAIR_UPSCALE_MODEL - the naming and the pairing pattern (a loader that picks a checkpoint, an apply node that runs it, single forward pass, no sampling loop) point to a classic single-pass GAN-style upscaler, the ESRGAN/Real-ESRGAN lineage that's been the workhorse of this category since 2021. That family is fast - milliseconds per image on a decent GPU, or effectively instant here since it's running on BizyAir's hardware anyway - and it doesn't hallucinate new detail. It sharpens and cleans up what's already there rather than generating texture that wasn't in the source.

That's a meaningfully different tool from the newer generation of diffusion-based restorative upscalers (SeedVR2, SUPIR, and similar) that have become popular for adding genuinely new fine detail - pores, fabric weave, that kind of thing - at the cost of being much slower and prone to occasionally inventing detail that wasn't there. If you're looking for that kind of aggressive detail generation and this node's results feel flat by comparison, that's not a misconfiguration - it's a different category of tool, and this pack (as of this writing) doesn't appear to expose that heavier style of upscaler under this particular node.

Inputs and outputs

• upscale_model - required, type BIZYAIR_UPSCALE_MODEL. This has to come from BizyAir's own upscale model loader node - it won't accept a standard ComfyUI UPSCALE_MODEL type from a local .pth file, because the actual model runs on BizyAir's servers, not yours.
• image - required, the image to upscale.

Output: a single IMAGE, upscaled by whatever factor the loaded model applies (typically 2x or 4x depending on which checkpoint you picked upstream - that choice happens in the loader node, not here).

How to install it

Ships with the full BizyAir pack.

ComfyUI Manager: search "BizyAir," Install, restart.

git clone:

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

Set your BizyAir API key before running - this node's execution, like every other node in the pack, happens server-side and needs an active account.

Common issues

"Model type mismatch" or the node won't connect. The most likely cause: you're trying to wire in a local ComfyUI UPSCALE_MODEL output (from a standard, non-BizyAir upscale loader) instead of this pack's own BIZYAIR_UPSCALE_MODEL loader. The two aren't interchangeable - find BizyAir's own upscale model loader node and use that.

Results look clean but not detailed. If you were expecting the kind of new-detail generation you've seen from diffusion-based upscalers elsewhere, you're likely hitting the ceiling of a classic GAN-style upscaler - see the category discussion above. This is expected behavior for this class of model, not a bug.

Large batches take longer than you'd expect for "fast" upscaling. Even though the model itself runs quickly, remember every call to this node is a network round trip to BizyAir's servers. Upscaling a big batch one image at a time will accumulate that overhead - factor that into how you structure a bulk job.

API key / credits. Same story as every BizyAir node - missing or exhausted credits surface as an execution failure on this node rather than a descriptive local error message.