How to Use the Builder

Step 0 — Pick a Media Type

The tabs at the top of the Builder (🖼️ Image, 🎬 Video, 🔊 Audio, 💬 Text) decide what kind of output you're prompting for. Switching tabs reloads the models, style categories, and suffix presets below to match — a Text tab won't show you camera lenses, and an Image tab won't show you "Reasoning Style."

Station 01 — Spill Your Brain

Main Idea — the big textarea at the top. This is the core scene, subject, or task. It's the only field that actually needs to be filled in for a useful prompt.

Sidekicks — click "+ Recruit Another" to add secondary subjects: extra people, objects, or details beyond the main idea. Each row has its own text, a Weight (0.1–2, see the cheat sheet below), and an optional Reference Image URL for models that can use one.

The No-Fly List — things you want kept out of the result (blurry, watermark, extra fingers, whatever). This only makes it into the actual request if the model you pick has a negative-prompt field; otherwise you'll see a warning instead.

Station 02 — Flavor Dust

This is the style library, grouped into collapsible categories (Medium, Camera, Lighting, Color, Artists, and more for images; Camera Movement and Visual Style for video; Mood and Genre for audio; Persona and Tone for text — the exact set depends on your media type).

Open a category and either type in the search box to filter its terms, or hit 🎲 Random to let it pick one for you. Click a term to select it — once selected, it gets its own Weight field using the same 0.1–2 scale as Sidekicks, so you can dial any style term up or down independently.

Weight Cheat Sheet

Both Sidekicks and selected terms share one weight scale. Here's what each range actually does to the assembled prompt text:

Exception: MidJourney doesn't use these phrases at all — it converts weight straight into native ::1.3-style multi-prompt syntax instead.

Pick Your Weapon

The Model dropdown in the sidebar lists every active provider/model for your chosen media type — MidJourney, ChatGPT, Higgsfield, Magnific, OpenAI, and others. Picking one shows its description right below, and unlocks the next panel.

Fine-Tune the Machine

Once a model is selected, "Fine-Tune the Machine" fills in with that model's actual parameters — things like aspect ratio, resolution, stylize/chaos sliders, or checkboxes like "tile" — pulled directly from that model's own schema. A different model means a different set of knobs here.

The Secret Sauce

The Preset dropdown appends a ready-made suffix to the end of your prompt (style boosters like "trending on artstation" for images, or instructions like "Force JSON Only" for text). Typing anything into Custom override replaces the preset entirely — it doesn't stack on top of it.

The Goods (Output)

Hit Refresh Preview — or the sticky ✨ Generate Prompt bar at the bottom of the screen — to assemble everything above into:

Prompt Text — the human-readable prompt, ready to paste into Discord, ChatGPT, or wherever it's headed. Request Payload (JSON) — the same content shaped into the fields that model's real API expects, for anyone calling it programmatically. Each box has its own Copy button. Check the Warnings box above them first — it flags things like a required field left blank, or a Do Not Include that got dropped because the model has no negative-prompt slot.

Stash It

Give it a Title and optional Notes, tick Favorite if it's a keeper, and hit Save to Library. Saved prompts show up on the Library page, where you can reload one back into the Builder, favorite/unfavorite it, or delete it.

Frequently Asked