🤖 Leon Pruna Image Upscale API

Before you reach for this node, figure out which of the three upscaling jobs you actually have. If your image is already sharp and you just need more pixels, a free ESRGAN or Lanczos pass locally is the right answer - faster, free, and it can't invent anything. This Leon_Pruna_Image_Upscale_Node is for the second job: generative upscaling, where a model adds plausible detail that wasn't there before. It calls Pruna's p-image-upscale model through the pack's HyprLab base, so all that detail generation happens on someone else's GPU and lands in your graph as a normal image.

That's the trade in one sentence: you give up local control (and per-call money) in exchange for zero VRAM and zero model management. If you're on a 6GB card and want 4K output from a small render, that trade can be very attractive.

How it works

Same plumbing as every API node in the pack - POST to https://api.hyprlab.io/v1/images/generations with a Bearer token, retry with exponential backoff (up to five attempts), decode the response into an RGBA tensor. The p-image-upscale payload takes your image, an upscale_mode, and either a target or factor value. Two ways to specify the size, and the node requires one of them to be used by the API:

• upscale_mode = target - upscale to a fixed multiplier. target of 4 = 4× upscale.
• upscale_mode = factor - same idea, different field. Set factor to 2 for 2×.

Both target and factor accept 1–8. The extra knobs, enhance_details and enhance_realism, are booleans you can stack - they toggle detail- and realism-enhancement passes in the API.

Inputs that matter

• image or image_url - the source. Feed either a ComfyUI IMAGE tensor (e.g. from a previous generation or LoadImage) or a hosted URL string. The code raises a ValueError if you provide neither. And don't wire both: the base class explicitly refuses "provide either an image tensor or an image URL, not both."
• upscale_mode / target / factor - pick your mode and set the matching value; the unused one is ignored.
• enhance_details / enhance_realism - the two toggles that actually change the character of the output.

Outputs are the pack standard: image, image_url, seed.

Installing

ComfyUI Manager (search Leon) or clone:

cd ComfyUI/custom_nodes/
git clone https://github.com/l3ony2k/comfyui-leon-nodes comfyui-leon-nodes
pip install -r requirements.txt

Restart, paste a HyprLab key. No downloads, no VRAM pressure - the whole upscale happens remotely.

Common issues

The "no input" error is the one beginners hit first: you must give it image or image_url, and the failure message tells you exactly that. After that, the generative-upscaling caveat from the wider community applies: a generative upscaler will rewrite faces and small details, so if you upscale a portrait, don't be surprised the identity drifts - the KB's upscaling guide flags this as design, not a bug. And because each call is paid, upscale the smallest source that still carries the detail you want; a 4× target on a tiny soft image is money spent inventing content, not recovering it.