The PixAI Prompt Formula | PixAI Mastery, Part 2
The 6-part formula for stable, high-quality PixAI prompts — Subject, Action, Environment, Style, Detail, Quality. Plus negative prompts and how to find tags that work.
📚 PIXAI IMAGE GENERATION MASTERY · 5-PART SERIES
Part 1: Model vs LoRA Foundations · Rookie
Part 2: The PixAI Prompt Formula ← you are here
Part 3: LoRA Stacking Guide · Rookie
Part 4: AI Art Composition Guide · Advanced
Part 5: Cinematic Lighting & Depth · Master
Models and LoRAs sorted. Now comes the question every PixAI user runs into next: what do I actually type in the prompt box?
The answer’s a little messier than you’d hope. Prompts come in two flavors. The classic format is tags — short, comma-separated keywords, the way most anime base models were trained to read. Tags are still the most reliable form for the majority of PixAI models, and good tag instincts come from practice.
But you’ve got shortcuts now. PixAI’s Prompt Helper lets you describe what you want in plain language and translates that into tags for you. Newer models like Tsubaki.2 read natural language directly — including non-English — and produce excellent results without any translation step. So no, you don’t have to memorize hundreds of tags before you can generate something good.
What stays constant across all of these formats is the structure. Tag format, plain English, Japanese, Chinese — the underlying recipe is the same. There’s a 6-part formula that consistently produces a stable, complete image. Once you can fill all six slots, you stop staring at a blank prompt box. You also stop getting outputs where the character is fine but the background is empty, or the scene is rich but the character is generic.
This article walks through the formula, then covers negative prompts (telling the model what not to draw) and artist tags (the secret-weapon style trick). By the end you’ll be writing prompts that produce what you actually imagined.
— THE FORMULA —
Six Slots, In Order
▸ THE 6-PART FORMULA
Subject + Action + Environment
+ Style + Detail + Quality
= a complete, balanced prompt
Each slot has a job. Order matters because PixAI’s models read prompts roughly in order of importance — what comes first gets weighted heavier. Let’s walk through each slot with a concrete example.
SLOT § 01 · SUBJECT
Who or what is in the image
The character or object — gender, basic appearance, defining traits. This is the most important slot because everything else builds on it.
Tag tip: Common counters like 1girl, 1boy, 2girls set the basic scene. Then add 2–3 distinctive features (hair color, eye color, age range).
SLOT § 02 · ACTION
What they’re doing, how they’re posed
Pose, expression, gaze direction. This brings the subject to life. Without an action you’ll get a stiff portrait that looks like a passport photo.
Tag tip: Combine a body action (standing, sitting, walking) with a face cue (looking at viewer, closed eyes) and an expression (smile, surprised).
SLOT § 03 · ENVIRONMENT
Where they are, what surrounds them
Setting, time of day, weather, season, key environmental elements. Skip this and you’ll get a character on a flat background.
Tag tip: Pick a broad setting first (outdoors, indoors, classroom, forest) then add 1–2 specific elements that signal the mood (a season, a weather, a memorable object).
SLOT § 04 · STYLE
The aesthetic register
The overall artistic feel — anime, illustration, watercolor, photorealistic. This shapes everything from line quality to color palette.
Tag tip: Most of the time the model itself already provides the dominant style — see Part 1. So you usually only need 1–2 style tags here as a nudge. Want a stronger style override? That’s what a Style LoRA is for.
SLOT § 05 · DETAIL
Lighting, atmosphere, finishing touches
Lighting type, atmospheric effects, depth-of-field, mood enhancers. This is what turns a competent image into one with a feeling.
Tag tip: This slot is where mood lives. Lighting tags (backlight, soft lighting, rim light) and atmosphere tags (depth of field, soft bokeh, floating particles) carry most of the weight. Part 5 is dedicated entirely to this.
SLOT § 06 · QUALITY
Resolution / craft markers
Generic quality boosters. These signal to the model that you want polished output. They go last because they’re modifiers, not content.
Tag tip: The classic trio is masterpiece, best quality, high detail. Don’t overdo this slot — five quality tags don’t beat three. And these tags cannot fix a broken composition. They polish. They don’t repair.
— PUTTING IT TOGETHER —
A Complete Prompt
Stack the six slots end to end and you have a working PixAI prompt. Here’s the full example:
▸ COMPLETE PROMPT
1girl, black hair, blue eyes, looking at viewer, slight smile, standing, outdoors, ginkgo leaves, autumn, anime style, soft illustration, soft lighting, cinematic light, depth of field, high detail, masterpiece, best quality
Read it slot by slot and you’ll see all six pieces. The result: a complete image, not a fragment. The character has a body, a pose, a place to be, a visual style, an atmosphere, and a polished finish. None of that would be there if any slot was empty.
The formula is also diagnostic. If your output looks empty in some way, identify which slot is thin or missing. No environment? Slot 3 is empty. Stiff character? Slot 2 is missing. No mood? Slot 5 needs work. The formula tells you what to fix.
— SECTION TWO —
Negative Prompts — Telling the Model “No”
Even with a perfect prompt, models slip in things you didn’t ask for. Extra fingers. A hat. A random second person standing in the background.
The fix is the Negative Prompt — a separate prompt box where you list things to avoid. You’ll find it tucked under the generator’s advanced settings.
Classic example: you ask for a male character and the model keeps generating women. The fix is to put 1girl in the negative prompt. Now the model knows: not female. Output corrects.
📌 STARTER NEGATIVE PROMPT
Don’t know what to put in your negative prompt? Good news: you don’t have to write one from scratch. Most PixAI models ship with a sensible default negative prompt baked in, so leaving the box empty is a perfectly fine starting point. Only step in when you’ve got a specific problem to fix — then add to (or override) the default with tags that target what you’re seeing. A typical beginner stack looks like this:
Add to it as you discover specific problems. If the model keeps producing busy backgrounds when you want a clean one, add cluttered background. If it adds glasses you don’t want, add glasses. The negative prompt is just as important as the positive one. Sometimes more.
— SECTION THREE —
Artist Tags — The Style Shortcut
One technique that punches way above its weight: artist tags. Tags that name a specific artist’s style — they pull the model’s output toward that artist’s aesthetic, the way a Style LoRA would.
Why this matters: anime base models often have a slightly “AI-feeling” sheen. Too smooth, too generic. Artist tags break that. They give the output a hand-drawn personality the model alone can’t produce.
You can stack multiple artist tags too. A chain of artist names blends their styles, the same way LoRA stacking blends multiple style LoRAs. This is sometimes called an “artist chain.” The first time you stack two artist tags and see your output suddenly look like it was actually drawn instead of generated, it changes how you write prompts.
📌 HOW TO FIND ARTIST TAGS
Most anime-trained models recognize artist names from Danbooru — a large public anime image board. Searching danbooru.donmai.us by tag will show you which artist names exist as tags. You can also browse the PixAI community feed: when someone posts an image with a strong style and the prompt is visible, the artist tags are usually right there. The most fluent users build a personal mental list of “artists I know what they do,” exactly the way they build a list of LoRAs.
— SECTION FOUR —
Don’t Know the Right Tag?
Beginners often hit a wall here: “I want a specific look but I don’t know what to type.” The honest answer involves three sources, used in order.
SOURCE 01
PixAI community feed
On the home feed, flip on the “Prompt Shared” filter — that narrows it down to images whose creators chose to make their prompts public. Find one you love, open it, and hit Use as Reference. The prompt, model, LoRAs, and settings all load straight into the generator. Tweak from there. This is the fastest learning loop on the platform — you start from a known-working setup and reverse-engineer why it works.
SOURCE 02
Danbooru tag system
For anime-leaning models, Danbooru’s tag taxonomy is basically the model’s vocabulary. If a tag exists on Danbooru, the model probably understands it. Search Danbooru for the visual element you want, see what tag it’s classified under, use that tag in your prompt.
SOURCE 03
The model’s own page
A lot of model creators publish recommended tags or trigger words on the PixAI model page. These are tags the model was specifically trained to respond to. Look here when a generic tag doesn’t seem to do anything.
— FAQ —
Common Questions
Should I write prompts in plain English sentences or as comma-separated tags?
Either works. The classic tag format is still rock solid — most anime-leaning base models were trained on Danbooru-style tag lists and respond beautifully to it. But the newer DiT-based models read natural language directly, in English or otherwise, and PixAI’s Prompt Helper translates plain descriptions into tags for you on the older ones. Use whichever feels more natural. A lot of users mix the two — natural language for the scene description, tags for the technical bits like lighting and quality.
How long should my prompt be?
For beginners, aim for 15–30 tags. Below 10 you’re underspecifying, and the model fills in random defaults. Above 50 — at the beginner stage — tags start fighting each other, and your intent gets diluted in the noise. The 6-slot formula naturally lands around 15–20 tags, which is a comfortable starting zone. Once you’ve got a feel for how individual tags pull on the output, all bets are off — advanced users routinely run much longer prompts (or much shorter ones) on purpose, and they know exactly why. Length is a starter rule, not a law.
What does (tag:1.3) mean?
That’s weight syntax. You can boost or reduce a tag’s influence by wrapping it. (red hair:1.3) says “emphasize red hair 30% more.” (red hair:0.7) reduces it. Useful when a feature isn’t appearing strongly enough, or is appearing too strongly. Keep weights between 0.5 and 1.5 — extreme values usually break the image.
My prompt looks complete but the output still misses what I want. What now?
Three places to check, in order. (1) The model — is it the right architecture for what you want? (2) The negative prompt — sometimes a missing thing is being suppressed by an over-broad negative. (3) Tag weights — boost the missing tag to ~1.3 and see if it appears.
Do I really need all six slots every time?
No — the formula’s a checklist, not a contract. For a complete, polished image, filling all six slots is the safest bet, and it’s why beginners are taught to do it. But once you know what each slot contributes, you can drop any of them on purpose. The point of the formula isn’t to fill every slot — it’s to know what each slot does, so you can choose. Use whatever subset gets you the image you want.
— READY FOR PART 3 —
Now Write Your First Real Prompt
Open the generator. Pick a subject. Run through the six slots in your head and fill them. Add a sensible negative prompt. Generate. Look at what came out, identify which slot is thin, fix it, regenerate. That’s the loop — and that’s how you actually learn prompts.
Continue the Series
PART 3 · ROOKIE · NEXT
Diagnose what your output is missing, then layer the right LoRA.
