BYOKey Magnific Upscale

Magnific is the upscaler that made "AI upscale" famous - the one that doesn't just sharpen, it adds detail. In the ComfyUI world, upscaling is normally a local affair: you'd reach for a fast ESRGAN or a heavier model depending on how much detail you want to invent. Magnific is the hosted, pay-per-call version of the same idea, and BYOKey_Magnific is how it joins your workflow: feed it an IMAGE, get a bigger IMAGE back, scaled up to 16x with a creativity dial that decides how much new detail the model invents.

Where this fits: as a final polish pass on a render you care about. The KB's upscaling framing splits local upscalers by how much they "invent" - and Magnific sits at the inventing end, which is exactly why it has a creativity slider and an optional prompt telling the model what kind of detail to add. If you've got a mid-tier render and want a premium-finish, big-budget look without buying better hardware, this is the shortcut.

How it works

It's one synchronous POST to https://api.magnific.ai/v1/upscale with your image (inline base64), scale, creativity, and optional prompt, using your key as a Bearer token. The API returns a JSON payload containing the upscaled image as base64, which the node decodes back into an IMAGE tensor. One call, one result, no polling. Note that unlike the rest of the pack, this node's endpoint is hardcoded - there's no base_url input, so relay/proxy support is not a thing here.

The inputs that matter

• image - the input IMAGE. This is required, and it's the whole point of the node.
• scale - 2, 4, 8, or 16. 2x is a clean, cheap resolution bump; 8x/16x are where Magnific's detail-invention shows up - and where the cost and the risk of artifacts go up.
• creativity - 0 to 1, default 0.5. This is the dial that controls how much new detail the model invents. Low values keep it faithful to the source; high values can add texture, grain, and structure that wasn't there. Start at 0.5 and tune per image - this is the difference between "upscaled" and "reimagined."
• prompt - optional text guiding what detail to add. Want "fine film grain" or "painted texture"? This is where you say so.

Output is a single IMAGE - take it straight to a preview or save node.

Installing it

It's one of 37 nodes in the comfyui-byokey pack, with no extra dependencies beyond ComfyUI's bundled torch/aiohttp/PIL/numpy/av. Install via ComfyUI Manager (search "byokey") or:

cd ComfyUI/custom_nodes
git clone https://github.com/MeteorAndy/comfyui-byokey.git

Restart, and it's under api/byokey/image.

Gotchas

Cost is the honest sting. This is a paid API where price roughly scales with output size - a 16x creative pass on a large image is genuinely expensive, so don't run whole batches through it. Also, upscaling is a one-way street: once Magnific invents detail at 8x, you can't get the faithful original back, so always keep the source. And high creativity on a human face or a detailed texture can drift into artifacts - that's the model inventing, and why the dial exists. The pack-standard hygiene note applies too: your key is stored in the widget and saved into the workflow JSON, so scrub it before sharing workflows. If you want that final premium finish on a few select renders, this is the node that delivers it.