Fal API Flux LoRA Config

This node doesn't generate anything. It's a config packet: two inputs, one output, and its entire job is describing a single LoRA well enough for fal.ai's servers to go fetch and apply it. Every LoRA-capable node in this pack - and there are several - takes its LoRAs through one or more of these, not as a direct field on the generation node itself.

The two inputs. lora_url is a plain string, and this is the one that actually trips people up: it has to be a direct, publicly reachable link to a raw .safetensors file, because fal's servers download it themselves with no browser session, no login, and no page-parsing behind them. A Hugging Face resolve/main/... link works. A CivitAI model page URL - the one you'd normally copy from your browser's address bar - almost never does, because that page serves HTML, not the file. If you're sourcing a LoRA from CivitAI, you need its actual download link, not the page you found it on. The README's own troubleshooting section flags exactly this: "make sure your LoRA URL is correct and compatible with FLUX.1 [dev]." scale is a float from 0.1 to 2, default 1 - the LoRA's effective strength. Most LoRAs are trained assuming a strength around 1, so 1 is a sane starting point; pushing meaningfully past 1.2–1.5 tends to look overcooked the same way over-strength LoRAs do in local ComfyUI workflows, more so the stronger the LoRA already is on its own.

The output is a single LORA_CONFIG, which plugs into any of the lora_1 through lora_5 slots on a compatible Fal API Flux node (every Dev-tier node in this pack except plain image-to-image and the ControlNet-only variants). You build one of these config nodes per LoRA you want active, and wire each into its own slot - there's no way to combine several LoRA URLs into a single config node, so five LoRAs means five of these on your canvas.

Why this matters specifically on Dev: FLUX.1 [dev] is the one Flux tier that's genuinely forgiving to train against, tolerating a messy single-subject dataset in a way the undistilled base model doesn't (it does fall apart past roughly three subjects crammed into one LoRA, for what it's worth) - which is exactly why the community LoRA ecosystem around Flux concentrates almost entirely on this tier, and why this pack only wires LoRA support into Dev-based nodes, never Pro.

Installing it. No node-specific setup - it ships with the rest of the pack. Install via ComfyUI Manager (search "ComfyUI-Fal-API-Flux") or clone https://github.com/yhayano-ponotech/ComfyUI-Fal-API-Flux.git manually and pip install -r requirements.txt. Every generation node downstream of this one still needs the pack's shared config.ini (a [falai] section with your fal.ai api_key) to actually run, even though this particular node makes no API call on its own.

Troubleshooting. If a LoRA seems to do nothing at all, test lora_url by pasting it into a fresh browser tab with no login active - if it doesn't immediately start downloading a file, fal's servers won't be able to fetch it either. If it's doing something but weaker or stronger than expected, that's scale, not the URL; nudge it in 0.1–0.2 steps rather than guessing.