Smart Parameter Help

1,200+ models, 1,000+ parameters — modelBridge helps you understand what’s what

fal.ai gives you access to over 1,200 AI models. Between them, those models expose more than a thousand different adjustable parameters — guidance scale, inference steps, denoising strength, LoRA weight, motion bucket, temporal consistency, safety tolerance, and hundreds more. No editor knows all of them. No editor should have to.

Most AI tools hand you a raw slider labelled guidance_scale, a range of 0–20, and nothing else — or they point you at API documentation written for machine-learning engineers. modelBridge takes the opposite approach: it assumes you’re a great editor, not an ML researcher, and explains each control in language that makes sense for the work you actually do.

The goal isn’t just to get you through today’s generation. It’s to teach you the craft over time — so the parameters that feel intimidating this week become creative tools you reach for on purpose next month.

Every control, explained right where you need it

Under each setting in Advanced Settings, modelBridge tells you what it is. For the controls that would genuinely trip up a video editor, a small ⓘ icon sits next to the label. Click it and an inline explanation opens with three things:

1. What it does — in plain language, tied to what actually changes in your output
2. Which way to move it — what happens when you go lower or higher, on or off
3. Where to start — a concrete number or range to begin from

Here’s what that looks like for one of the most-asked-about settings:

Guidance Scale ⓘ How closely output follows your prompt. Lower = more creative, higher = more literal. Start ~6–8. Learn more ↗

Compare that to the raw description the model itself ships with — “The CFG (Classifier Free Guidance) scale is a measure of how close you want the model to stick to your prompt” — which is accurate but assumes you already know what CFG is. modelBridge keeps the model’s own wording available and layers a plain-language explanation on top for the controls that need one.

A few more, exactly as they read in the panel:

Strength ⓘ — How much the image changes. Lower = keeps your original, higher = more new content. Start ~0.5.

LoRA Scale ⓘ — How strongly the style add-on applies. Start at 0.7. Lower = subtle, higher = more pronounced.

Inference Steps ⓘ — Detail vs. speed. Lower = faster, higher = cleaner. Start ~20–30.

Guidance where it matters — and nowhere it doesn’t

The ⓘ icon is a promise: if it’s there, there’s something worth reading. Self-explanatory controls — Duration, Resolution, Aspect Ratio, Number of Images — don’t get one, because they don’t need one. Around 300 of these obvious fields are deliberately left clean, so the icon never becomes noise you learn to ignore.

And there are no dead ends. Every parameter modelBridge shows you leads to a real explanation — over 700 of them hand-written specifically for editors and motion designers, spanning the entire fal.ai catalog: video generation, image synthesis, voice and audio, upscaling, inpainting, ControlNet, LoRA, and more. You’ll never click ⓘ and land on “see API docs.”

Across the catalog that’s 1,020 distinct parameters, and modelBridge has already sorted every one of them — written a tip for the ones that help you, and hidden the icon on the ones that don’t. Whenever fal.ai adds new models, their new parameters get the same treatment before they reach you.

The deeper educational material: the Parameter Reference

Every Learn more ↗ link opens the modelBridge Academy Parameter Reference — the companion to the in-panel tips, and the place to go when you want to genuinely understand a setting rather than just grab a starting value.

It covers over 100 parameters in depth, organised into categories: prompt & guidance, generation quality, LoRA & style, video & animation, image quality, audio, ControlNet, IP Adapter, and more. Each section is written the same way:

• In short — the one-line version
• What it does — the real explanation
• How to think about it — using Premiere Pro and After Effects analogies you already know (guidance scale is like directing an actor; LoRA scale is like the opacity of a LUT; inference steps are like render passes)
• Recommended settings — concrete ranges for low / default / high, and when to use each
• Common mistakes — the ones editors actually make
• Also called — the other field names the same setting hides behind across different models

If you want the short path, start with The 4 parameters that control every generation: guidance scale, inference steps, seed, and strength shape almost every result you’ll ever make.

Built to be outgrown

The whole system exists to make you more capable, not more dependent. The tips are guidance, not guardrails — and once a parameter is second nature, you can turn Show learning tips off from Settings to keep the panel clean.

That’s the intent. The first time you meet a model, you always know what a control does, which way to move it, and where to start. The hundredth time, you already know — and the help stays out of your way. The best creative tools don’t just hand you controls; they teach you to master them.