Clarity Upscaler (fal)

Despite the generic name, this is the Clarity upscaler - philz1337x's open reimplementation of Magnific AI, a MultiDiffusion + ControlNet + LoRA stack. Its whole personality is the two knobs creativity and resemblance, which is the giveaway: this is a creative upscaler that reimagines detail into your image, not a faithful one. Think of it as the opposite end of the ladder from SeedVR2. This node runs it on fal.ai.

Cloud call, per-image billing, no local setup. Which is nice, because a MultiDiffusion tiled upscale is exactly the kind of thing that's fiddly to run yourself.

How it works

Clarity tiles your image, runs diffusion on each tile with a ControlNet keeping tiles anchored to the source, and blends them back together - adding detail as it goes. The amount of invention is yours to set. Crank creativity and it hallucinates fresh texture and detail Magnific-style, which looks stunning on landscapes and abstract material and terrible on a face you recognize. Keep it low and lean on resemblance and it behaves. This is design intent, not a bug: you're trading fidelity for "wow."

The inputs and outputs that matter

The two that define the node:

• creativity (default 0.35, 0–1) - how much new detail it invents. Higher = more dramatic, more hallucination, more drift from the original.
• resemblance (default 0.6, 0–1) - how tightly the result sticks to the source. Higher holds the composition; it's the counterweight to creativity.

The supporting cast:

• image (required) and upscale_factor (default 2, 1–4 in 0.5 steps).
• num_inference_steps (default 18) and guidance_scale (default 4) - standard diffusion dials; the defaults are sane.
• negative_prompt (defaults to the classic (worst quality, low quality, normal quality:2)) - steer it away from junk.
• enable_safety_checker and an optional seed (default -1).

Output is a native IMAGE.

How to install it

From gokayfem's ComfyUI-fal-API pack. ComfyUI Manager: search ComfyUI-fal-API, install, restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/gokayfem/ComfyUI-fal-API.git
pip install -r ComfyUI-fal-API/requirements.txt

Add your fal key (copy config.ini.example → config.ini, paste from fal.ai/dashboard/keys, or export FAL_KEY=your_key), restart.

Common issues & troubleshooting

It changed the face / added details that weren't there. That's creativity doing its job. This upscaler invents - it's the Magnific behaviour. Drop creativity toward 0.2 and raise resemblance toward 0.8 if you need the image to stay itself. For a recognizable person, upscale the face on a separate, low-creativity pass or keep it out of this one entirely.

Wrong tool for a clean image. If your source is already sharp and you just want more pixels and fine detail without reinterpretation, SeedVR2 (or a plain ESRGAN pass) is the better call. Clarity earns its place when you want the image reimagined richer, not preserved.

Result looks mushy or over-smoothed. Nudge creativity up a little and check your upscale_factor isn't asking for a bigger jump than the source can support.

Insufficient funds / billing lag. fal is pay-per-call and legit, but top-ups have sometimes posted late. Wait for credits rather than repaying. fal's errors surface as readable text in ComfyUI.