Midjourney Personalization in Practice: Train and Apply Your Aesthetic Profile
Hands-on Midjourney Personalization: build your aesthetic profile under V8.1, run A/B comparisons with --stylize, then deliver finished images with your personal look from SD through Run as HD.
This tutorial walks you through a complete Midjourney Personalization exercise: first clarify how it differs from SREF / Moodboards, then run “profile off / on” and “low / high --stylize” comparisons, and finally deliver finished images with your personal look via SD→Run as HD.
Best if you already know basic generation. If you have not started yet, begin with Getting Started with Midjourney: Your First Complete Generation Guide. For the full parameter list, see Midjourney V8.1 Parameter Cheatsheet: –hd –raw –chaos –stylize; for style anchoring, read the blog Midjourney Style Reference Complete Guide: How SREF and Moodboards Keep Series Visuals Cohesive.
What Personalization Solves
Plain prompts are good at saying “what this image shows,” but weak at guaranteeing “this is my aesthetic.” Common symptoms:
- The same person generates today leaning illustration, tomorrow leaning studio photography—no stable “me”
- You want a personal brand / account look but rewrite the mood with adjectives every time
- Style reference is on, yet results still feel like “someone else’s nice image, not mine”
Personalization compresses your long-term aesthetic preferences into a callable profile. Once enabled, together with an appropriate --stylize, the model is more willing to package the image to your taste—that is the core switch in this lesson.
How Personalization Fits with Other Style Tools
| Tool | Best for | What you mainly get | Notes |
|---|---|---|---|
| Personalization | Long-term personal aesthetic profile | Overall mood that feels “like me” | Often more visible with higher --stylize |
| SREF / Moodboards | Project / brand series look | Unified anchor that feels “like this set” | Manages project mood, does not replace personal profile |
| Image reference | Shape, pose, composition | Controllable structure | Handles “what it looks like,” not specifically “whether it looks like me” |
--raw |
Realism, less filtering | Weakens default beautification | Can stack with Personalization, but change only one variable at a time |
In one line: Personalization handles “who I am,” SREF / Moodboards handle “what this project looks like,” image reference handles structure, and text handles what to change in this image.
What You Will Learn
- Decide when to enable Personalization and when to turn it off for comparison first
- Run an A/B with the same prompt: profile off vs on
- After enabling the profile, use low / high
--stylizeto see how strongly the mood intervenes - Explore in SD under Midjourney V8.1 and deliver roughly 2K HD via Run as HD
- Know when to stack SREF and when to hold off
Estimated time: 25–35 minutes.
Before You Start
- Midjourney workspace account (this site’s “Try Midjourney”)
- Confirm the model is V8.1
- Your account has a usable Personalization profile (if the platform requires preference training / rating first, complete that flow)
- One simple subject (recommended: portrait by a window / tabletop still life / street corner scene—pick one)
- A notepad: record “off / on,” stylize values, and perceived differences
If your profile is still “empty” or just started training, this lesson still works for on/off comparison; differences grow clearer as the profile matures.
Exercise Overview
| Step | What to do | Success criteria |
|---|---|---|
| 1 | Write your “three-sentence mood” | Can state preferences verbally without stacking style words |
| 2 | Baseline with Personalization off (SD) | One “content-correct” comparison image |
| 3 | Same prompt with Personalization on | Can point out mood differences (color temperature, materials, atmosphere) |
| 4 | Low --stylize vs high --stylize |
Can describe how strongly the profile intervenes |
| 5 | Optional: add only one SREF / reference and compare | Know which layer dominates after stacking |
| 6 | Small-step iteration (Vary / Remix) | Mood stays stable, content gets more accurate |
| 7 | Run as HD final and record | ~2K delivery + parameter notes |
Step 1: Write Your “Three-Sentence Mood” First
Before opening the profile, skip prompt-stacked styles and write three lines:
- Medium preference: e.g. “lean photographic, slight film grain, no heavy illustration brushwork”
- Color preference: e.g. “low-saturation cool-warm neutrals, no neon”
- Taboos: e.g. “no over-smoothed skin, no flare explosions”
These three lines serve two purposes: checking whether results feel “like you,” and judging whether differences come from the profile or from the prompt itself.
Step 2: Turn Personalization Off—Baseline (SD)
Turn Personalization off and generate in SD with a short prompt:
a person sitting by a rainy window, soft daylight, quiet mood, photorealistic --stylize 100
Pick one image where content direction is correct—subject clear enough. This is your no-profile comparison sample.
Use SD in exploration to spend budget on “information” not “pixels.” See the blog Midjourney HD vs SD: Save Time and Cost with a Resolution Strategy.
Step 3: Same Prompt with Personalization On (A/B)
Keep prompt, --stylize, and model unchanged—only turn Personalization on—and generate another round.
When comparing, answer:
- Is color temperature / contrast closer to your “three sentences”?
- Do materials and atmosphere feel more “like images you usually pick”?
- Is subject information still readable, or drowned by mood?
If you barely see a difference: check whether the profile is trained and truly enabled; or temporarily raise --stylize slightly and compare again (the next step practices this systematically).
Step 4: Low vs High --stylize (Profile On)
With Personalization on, --stylize often decides how deeply the profile intervenes. Run two comparisons (everything else unchanged):
A. Lower (near default)
a person sitting by a rainy window, soft daylight, quiet mood, photorealistic --stylize 100
B. Higher (let the profile drive mood more)
a person sitting by a rainy window, soft daylight, quiet mood, photorealistic --stylize 250
(Adjust values to your workspace default and range; the point is same prompt, only stylize changes.)
Observe:
| Phenomenon | Possible meaning | Suggestion |
|---|---|---|
| Higher stylize feels more like your taste | Profile is working | Record this “sweet spot” range |
| Higher stylize blurs subject, loses information | Aesthetic intervention too strong | Lower stylize, or clarify subject in prompt |
| High and low look almost the same | Weak profile / not active / prompt already very stylistic | Check switch and training; or try a more neutral prompt |
For parameter meaning, revisit Midjourney V8.1 Parameter Cheatsheet: –hd –raw –chaos –stylize.
Step 5 (Optional): Stack Only One Other Constraint
Once profile and stylize are stable, add one constraint as needed (do not add everything at once):
| Need | Add | Tutorial / guide |
|---|---|---|
| Unified project series look | SREF / Moodboards | Series Visuals with Style Reference: Brand Consistency in Midjourney |
| Shape / pose follows reference | Image reference | Midjourney Image Reference in Practice: Controlled Variants from Reference Images |
| More realism, less filtering | --raw |
Midjourney –raw Realism Tutorial: Product Still Life & Architecture Step by Step |
Stacking rule: change only one switch at a time. If “like me” and “like the project” clash, turn one off for A/B first, then decide which to keep.
Step 6: Small-Step Fixes with Vary / Remix Until Deliverable
Once direction and mood are both right:
- Details slightly off → Vary Subtle
- Composition wrong → Vary Strong (watch you do not drift outside your “three sentences”)
- Description gap → Remix, change only one sentence at a time
Full iteration lesson: Midjourney Vary & Remix: Small-Step Iteration to Final Delivery. Stay in SD throughout; HD only after satisfied.
Step 7: Run as HD Final and Delivery Check
- Apply Run as HD or
--hdto the final winning result (roughly 2K HD) - Check: does it still match your “three-sentence mood”?
- Check: is subject information clear and readable?
- Save notes: Personalization on/off,
--stylizevalue, whether SREF / reference was stacked, final prompt
V8.1 / V8.2 do not support
--q. Use SD→HD for exploration and delivery—do not carry over V7’s--q 2/--q 4.
In-Class Exercises: Three Comparisons (Do These)
Exercise 1: Profile On/Off
Same prompt, same stylize → Personalization off / on → write one sentence describing the mood difference.
Exercise 2: High vs Low Stylize
Profile on → change only --stylize high/low → note your “sweet spot” value.
Exercise 3: One Stack at a Time (Optional)
On the winning setup, add only SREF or image reference or --raw → judge whether it still feels “like you.”
Complete everything in SD; Run as HD only on the most satisfying image at the end.
Decision Cheatsheet (Use Next Time)
| Your goal | Start here |
|---|---|
| Stable “my flavor” | Enable Personalization and try raising --stylize |
| Unified project / brand look | Prioritize SREF / Moodboards; profile can come later |
| Accurate shape / structure | Image reference; do not crank profile at the same time |
| No difference after toggling | Check training and switch; try neutral prompt; test higher stylize |
| Mood right but details off | Vary / Remix, one variable at a time |
| Already satisfied | Run as HD, stop stacking switches |
FAQ
Do Personalization and SREF have to be on together?
No. Personal content can use only the profile first; brand series can use only SREF / Moodboards first. They can stack, but when problems appear, split them for comparison.
Is higher --stylize always better?
No. Too high can drown subject and scene information. This lesson’s goal is finding the range that works for your profile, not chasing the maximum.
Will the profile make every image look the same?
There will be “family resemblance”—both a strength and a risk. Change subjects, change lenses, turn the profile off for comparison when needed, so outputs do not all become the same filter.
Should I enable Personalization for realism tasks?
You can, but for neutral product / architecture reference, try --raw first or lower stylize to avoid heavy personal filtering. See Midjourney –raw Realism Tutorial: Product Still Life & Architecture Step by Step.
What to learn next?
- Series look: Series Visuals with Style Reference: Brand Consistency in Midjourney
- Iteration to final: Midjourney Vary & Remix: Small-Step Iteration to Final Delivery
- End-to-end workflow: blog Midjourney AI Art from Start to Finish: The Complete 2026 Workflow (V8.1)
Once your subject is ready, open the workspace from “Try Midjourney” at the top of the page and run all seven steps for your first round of “train and apply your aesthetic profile.”
Summary
The value of Midjourney Personalization is turning “occasionally like me” into a callable long-term aesthetic. Write your three sentences first, then run off/on and high/low stylize comparisons, stack only one other anchor when needed, and deliver via SD→Run as HD. Build the habit of one variable at a time—your personal look will stabilize instead of relying on luck to hit the right filter each time.