Superside Juggernaut Flux Pro Image-to-Image

Juggernaut is a name the SD community knows - a checkpoint lineage famous for pushing photorealistic output as far as local models can take it. This node hands you the "Pro Flux" variant of that family as a hosted image-to-image call on fal.ai, which means you get the Juggernaut look without the multi-gigabyte download or the VRAM bill. The cost, as always with this pack's API nodes, is per-call and your image leaves the machine.

So what is it for? Re-styling an existing image into high-realism territory. You drop in a photo or a render, write what you want it to become, and strength decides how much of your input survives. Think: turning a decent render into something that passes for a photo, or giving a flat product shot cinematic photographic lighting. The model runs entirely on fal's hardware - this is the "the model isn't on my machine" branch of the API-node decision tree, not the "no local option" one, since Juggernaut's Flux-lineage weights are open and runnable locally if you have the card for it.

The inputs that matter

• image, prompt, api_key - the essentials. Prompt it like a photo brief; Juggernaut responds well to photographic vocabulary.
• strength (default 0.95) - the big one. This plays the same role as a sampler's denoise: 0.95 means "heavily transform, keep the composition," while values under ~0.5 drift toward the prompt's interpretation of the image. The default is aggressive - if the result barely resembles your input, that's the setting to pull back.
• guidance_scale (default 3.5) - how closely it follows the prompt. Flux-lineage models like lower CFG than SD 1.5 habits suggest; 3.5 is already near the sweet spot.
• num_inference_steps (default 40, 10–50) and seed - steps are the usual trade of quality for time; set a seed if you want reproducible results.
• num_images (1–4) - variations per run. Each one is a separate billable generation, so don't set 4 out of habit.
• enable_safety_checker (on by default).

The single output is the transformed IMAGE, ready to save or continue down the graph.

How it works

The same Superside machinery you'll see across this pack: upload your image to fal, call the Juggernaut Flux Pro img2img endpoint with strength/CFG/steps, download the result back as a tensor. No local model, no weights, no VRAM. Because strength is a per-run dial, expect to tune it per input image - a composition you love will want a lower value than one you're happy to let the model reinterpret.

Installing it

It ships with the Superside pack:

cd ComfyUI/custom_nodes
git clone https://github.com/Superside/comfyui-superside-nodes
pip install -r requirements.txt

Restart ComfyUI and find it under Superside. You'll need a fal key pasted into api_key (no config file; blank key falls back to a FAL_KEY env var).

Where people get burned

Cost is the first thing - at 40 steps with variations enabled, a few test runs become a real line item, so tune on num_images = 1 and a lower step count, then commit. The second trap is strength. Coming from local img2img habits, people leave it at 0.95 and then wonder why their brand logo or character came back changed; if the task is restyle but keep the identity, 0.6–0.7 is a saner starting point. And keep the photorealism expectations honest: Juggernaut-style models make images that read as photos - which is exactly why the small things, faces and text and hands, deserve a close look before you call a run done.