PVL FLUX DEV LORA (fal.ai)

You trained a Flux LoRA - on your character, your art style, your product - and now you want to generate with it, but loading FLUX.1 Dev plus a LoRA locally needs VRAM you may not have. PVL FLUX DEV LORA runs the job on fal's flux/lora endpoint instead: your prompt, your LoRA identifier, a few sampler settings, and fal does the inference and sends the image back. If the pack's plain Flux Dev node is the "just generate" option, this is the "just generate, with my LoRA" option.

It's part of the ComfyUI Assistant Node pack from pvlprk. One important thing to understand before you start: the LoRA names here are fal-side references - paths to models you've uploaded to fal (or fal-hosted models), not local .safetensors files. You upload your LoRA to fal's storage and give this node the identifier.

How it works

Text-to-image with up to three LoRA stacks, each a name + scale pair (scale range −2 to 2, default 1.0). The node submits each call through fal's queue API in two phases - submit, then poll until completion - and downloads the result as an IMAGE tensor. It'll also split a multi-part prompt on the delimiter (default [++]) and map those parts onto your requested num_images, reusing the last prompt if you ask for more images than prompts.

The sampler side is FLUX.1 Dev standard: steps (default 28), CFG (default 3.5), width/height (256–1440, default 1024×1024), seed, num_images (1–4), output_format, and sync_mode. Plus the pack's reliability trio: retries, timeout_sec, and debug_log. Content-policy violations are detected and skipped rather than retried.

The inputs that matter

• prompt - your generation prompt, tuned for the LoRA.
• lora1_name / lora1_scale - the fal-side LoRA path and its strength. The lora2_* and lora3_* slots stack additional LoRAs.
• steps / CFG - defaults 28 / 3.5.
• num_images - 1–4, and it works with the delimiter for batched prompts.

Output: one IMAGE.

Installing it

Part of pvlprk/comfyui-pvl-api-nodes. ComfyUI Manager (search "ComfyUI Assistant Node") or:

cd ComfyUI/custom_nodes
git clone https://github.com/pvlprk/comfyui-pvl-api-nodes

Set FAL_KEY (or FAL_API_KEY / FAL_CLIENT_KEY), restart, and have credits.

Where people get burned

The #1 mistake is typing a local path into lora1_name and wondering why it fails - this node can't see your models/loras folder; the name must resolve on fal's side. Check the console with debug_log on to see exactly what identifier went out. Also, LoRA strength isn't free: pushing scale past ~1.2 can start overriding the base model's natural-language understanding, and negative scales are real but weird - if you go there, know what you're testing. Standard FAL_KEY and blank-placeholder gotchas apply.