---
title: "Getting started with Higgsfield"
description: "A first-session orientation: what to try first, the two decisions that shape every generation, and the mistakes that waste the most credits early on."
url: "https://higgsfield.wiki/guides/getting-started/"
verified: "2026-08-26"
publisher: "Higgsfield Wiki — independent reference, not affiliated with Higgsfield AI"
---

# Getting started with Higgsfield

A first-session orientation: what to try first, the two decisions that shape every generation, and the mistakes that waste the most credits early on.

A short orientation for a first session, aimed at getting a usable result rather than at touring every feature.

## The two decisions that shape everything

Before touching a prompt box, settle these. Almost every early frustration traces back to getting one of them wrong:

  - **Image or video?** Video costs far more per generation. If you are unsure what you want, work it out in images first and animate afterwards.
  - **Does something specific have to appear?** A real product, a real face, a layout you designed. If yes, you need a reference image — description cannot carry identity. If no, a text prompt is fine.

Those two answers pick your mode: [text to image](/text-to-image/), [image to image](/image-to-image/), [text to video](/text-to-video/), or [image to video](/image-to-video/).

## A first session that teaches you something

  - **Generate one image from text.** Name subject, setting, lighting, framing and one style word. Note what it ignored.
  - **Change only the lighting** and run again. This single comparison teaches more about prompting than any guide.
  - **Feed a result back in** through image-to-image and change one thing. Now you understand the controllable mode.
  - **Animate an approved still.** One action, one camera move.
  - **Watch it full screen.** Judge stability before anything else.

Five generations, and you will have a working model of what the tools do and where they fail.

## What wastes credits early

| Mistake | Instead |  |

| Starting with video | Settle the look in images; they cost a fraction |  |

| Describing a specific person or product | Supply a reference image |  |

| Rewriting the whole prompt between runs | Change one variable so you learn from the comparison |  |

| Stacking "8K, masterpiece, ultra detailed" | Describe the actual light and lens |  |

| Prompting a sequence of events | One action per clip |  |

| Judging on a phone preview | Full resolution, full screen |  |

## Where to go next

For the tool families and how they differ, [what is Higgsfield](/what-is-higgsfield/). For choosing a model, the [model reference](/models/). For terminology, the [glossary](/glossary/). For the full video sequence, [how to make an AI video](/how-to-make-ai-videos/).

## Common questions

### What should I generate first?

One image from a text prompt, then the same prompt with only the lighting changed. That comparison teaches more than any amount of reading.

### Why is my output nothing like what I imagined?

Usually a vague prompt or the wrong mode. If something specific had to appear, description alone will not produce it — supply a reference image.

### How do I avoid wasting credits?

Work out the look in images before spending on video, and change one variable at a time so each generation teaches you something.