Getting started: your first AI image prompt

What Prompt Builder does (and doesn't)

Prompt Builder is a prompt workspace, not an image generator. It helps you write, organise and fine-tune the text prompt that an AI model turns into a picture. The image itself is rendered somewhere else — on your own computer or an online service. That separation is deliberate: it keeps Prompt Builder free, private (your prompts never need to leave your device), and usable with any model — Stable Diffusion, SDXL, Pony, Flux, Midjourney and more.

So the flow is always two halves: (1) build the prompt here, then (2) generate the image in a model. Let's do both.

Part 1 — Build a prompt

1. Pick a model (and LoRA). At the top of the Build pane, choose the model you'll generate with. Snippet buttons that only apply to a specific model or LoRA appear once it's selected; leave the dropdowns on “All” to see everything.
2. Stamp in snippets. The left pane is full of reusable macro buttons grouped by category (quality, subject, lighting…). Click one and its text is inserted into the live editor at your cursor, with separators added automatically. Type your own words anywhere too.
3. Tune attention weights. Select a word or phrase and press Ctrl/Cmd + ↑ to emphasise it (it becomes (word:1.1)) or Ctrl/Cmd + ↓ to reduce it. See the weighting & LoRA syntax guide for the details.
4. Add a negative prompt. The lower box is for things to avoid — worst quality, low quality, blurry, extra fingers. Buttons with a red dot route their text here automatically.
5. Copy it out. Use Copy Positive or Copy Both in the footer to grab the finished prompt, ready to paste into a generator.

Part 2 — Generate the image

You have two routes, depending on whether you want zero setup or full control and no limits.

Option A — Online, no install (easiest)

Paste your prompt into a browser-based generator. Many run Stable Diffusion or SDXL with a free tier and nothing to install — good for trying things out on any device, including a phone. Popular options include hosted Stable Diffusion playgrounds and community model sites (for example services built around Hugging Face Spaces, or the generators on civitai-style model hubs). Search “free Stable Diffusion online” and paste your positive and negative prompts into the matching boxes. Free tiers are usually rate-limited and queue at busy times.

Option B — Local, on your own GPU (full control)

For unlimited, private, uncensored generation you run the model yourself. The two most common backends are Automatic1111 / Forge (a friendly web UI) and ComfyUI (a node-based pipeline). Both are free and open-source; you download a checkpoint model (e.g. SDXL or Pony) and run it on a reasonably modern GPU. Our ComfyUI & Automatic1111 guide explains how prompts work in each.

Once a backend is running locally, Prompt Builder can talk to it directly — open the in-app generation panel, point it at your A1111 or ComfyUI address, and send the current prompt without copy-pasting.

A note on hardware

Local SDXL is comfortable on an NVIDIA GPU with ~8GB VRAM or more; less and you'll lean on smaller models or the online option. No capable GPU? Cloud notebook runtimes and rented-GPU services can run Automatic1111 or ComfyUI for you — a middle ground between the two routes above.

Reuse a prompt from an existing image

Already have a PNG you (or someone) made in Automatic1111 or ComfyUI? Drag it onto the editor — Prompt Builder reads the metadata baked into the file and recovers the exact positive and negative prompts, so you can study, tweak and re-run it.

Where to go next

• Browse the style guides for copy-ready prompts you can open straight in the builder.
• Learn the syntax in the weighting & LoRA guide and the glossary.
• See model-specific advice for SDXL, Pony or Flux.
• Share your best setups and discover others in the community.

Frequently asked questions