FalFluxLoraAPI

FalFluxLoraAPI is the node that makes the cloud-Flux idea actually useful, because it's the one that understands LoRAs. It runs fal-ai/flux-lora (or the image-to-image variant, if you feed it an image), so you can apply a character or style LoRA to a cloud Flux generation without ever loading a model locally. LoRAs are the small adapter files that teach Flux a specific subject or style - the cloud equivalent of dropping a .safetensors into your LoraLoader, except the model it's patching lives on fal's servers, not your disk.

This is also the node the author added in direct response to a feature request on the pack's launch thread: someone wanted to attach LoRA URLs to fal Flux, and here we are.

How it works

The LoRA plumbing is the interesting part. You don't type a filename - you feed this node a URL to a LoRA file plus a scale, and it gets those via a JSON string. That string comes from the pack's helper node, FalAddLora, which is a pure string-builder: it takes a lora_url and scale, appends {"path": lora_url, "scale": scale} to a JSON list, and hands the result to this node's loras input. Chain several FalAddLora nodes together (the loras output of one feeds the optional loras input of the next) to stack multiple LoRAs, then terminate the chain here.

If you connect an image to the optional input, the node switches to the flux-lora/image-to-image endpoint and i2i_strength comes into play, mirroring FalFluxI2IAPI's behavior - including the automatic downscale to 1024px unless no_downscale is set.

The inputs that matter

• loras - the JSON string from FalAddLora. This is the only thing that must come from somewhere else; the node won't do anything useful if it's empty.
• prompt - as always with Flux, natural language.
• width / height - 1024² defaults, 16px increments.
• steps - 25 default, tuned for dev.
• cfg - guidance, 3.5 default.
• i2i_strength - only relevant when an image is connected.
• no_downscale - skip the 1024px input cap.
• image_count - 1 to 4 outputs as a batch.

Output is IMAGE, batched if image_count is up.

Install

Manager → search "ComfyUI-Cloud-APIs", or clone into custom_nodes, restart, drop a fal key into keys/. Full drill:

cd ComfyUI/custom_nodes
git clone https://github.com/BetaDoggo/ComfyUI-Cloud-APIs

Nothing heavy to download - the whole pack's deps are fal-client, replicate, requests, numpy, pillow.

Troubleshooting

• Nothing happens / "loras" error - the loras input expects valid JSON from FalAddLora. Don't type a URL straight into it; wire the helper in.
• LoRA doesn't seem to apply - check the scale. Cloud LoRA scaling follows the same rules as local: 1.0 is the default but often too strong; 0.5–0.8 is a common sweet spot.
• LoRA + i2i fighting you - Flux i2i is delicate, and adding a LoRA makes it more so. Start with i2i_strength near 0.9 and adjust down carefully.
• Cost - every generation is a billed fal image, and LoRA + i2i is no exception. Iterate cheap (fewer steps, smaller size) before committing.

The pack is archived, but again, it's a thin API wrapper - it keeps running until fal changes something.