Midjourney V7 dropped in April 2026, and honestly, it changed everything for portrait work. The combination of better prompt understanding and the style reference system (—sref) means you can finally generate portraits that look like they came from a professional shootnot an AI experiment gone sideways.
I’ve spent months testing V7 portrait workflows. What follows are the prompts that actually work, the SREF codes that produce consistent results, and the troubleshooting fixes for when things go sideways. No fluff, no filler.
What Is the Midjourney —sref Parameter?
The —sref parameter is Midjourney’s style reference tool. It lets you apply the visual aesthetic of any imageor a numeric style codeto your generations. Think of it as handing Midjourney a mood board and saying “make everything look like this.”
You use it two ways: upload a reference image URL with --sref [url], or grab a numeric style code from Midjourney’s library with --sref [number]. The model then borrows the colors, lighting, texture, and overall vibe from that reference while you control the subject.
The style weight parameter —sw controls how strongly the reference influences output. Range is 0 to 1000, with 100 as the default. Higher values lock in style harder; lower values give your prompt more breathing room.
According to Midjourney’s official documentation, —sref works with versions 6 and 7, and the V7 model gives broader, richer style transfer than V6. If you’re still on V6, you’re missing out.
Midjourney V7 vs V6 for Portrait Work
| Feature | V6 | V7 |
|---|---|---|
| Prompt accuracy | Good | Excellent |
| Skin texture realism | Solid | Significantly improved |
| —sref style transfer | Functional | Broader, richer |
| Eye detail | Acceptable | Near-photographic |
| Hand generation | Problematic | Still challenging |
| Default —sv version | N/A | 6 |
V7 became Midjourney’s default model on June 17, 2026. If you’re still running V6 prompts, switch over. The portrait quality difference is substantial.
V8 alpha launched March 17, 2026, with V8.1 following April 14, 2026. But V7 remains the workhorse for photorealistic portraiture until V8 stabilizes.
Top 10 Photorealistic Portrait Prompts with SREF Codes
These prompts are tested and working in V7. Each includes a recommended SREF code range for portrait photography styles.
Prompt 1: Dramatic Studio Portrait
Prompt: “Photorealistic portrait of [subject], dramatic Rembrandt lighting, dark background, catch lights in eyes, visible skin pores, 85mm lens bokeh, fashion magazine quality —ar 3:4 —v 7 —sref 1772454150 —sw 300”
This is your go-to for professional headshots. The SREF code 1772454150 produces that clean, high-contrast editorial look. Rembrandt lighting creates that classic triangle under the eyes that makes faces pop.
Prompt 2: Environmental Portrait
Prompt: “Photorealistic environmental portrait of [occupation], subject in authentic work context, natural ambient light mixing with practical sources, documentary photography aesthetic —ar 3:4 —v 7 —sref 3357887304 —sw 250”
Environmental portraits tell stories. The subject matters more than perfection hereaim for authenticity over polish. Use a lower style weight (250) so the documentary feel doesn’t get overproduced.
Prompt 3: Golden Hour Outdoor
Prompt: “Photorealistic portrait of [subject], late afternoon golden hour light raking across face, intentional lens flare, shallow depth of field, warm color temperature, catchlight from sun reflected in eyes —ar 3:4 —v 7 —sref 3182173932 —sw 350”
Golden hour flatters every skin tone. The warmth hides imperfections and creates atmosphere that flat lighting can’t match. This SREF code pulls that warm, cinematic sunset aesthetic.
Prompt 4: High Key Beauty Portrait
Prompt: “High key beauty portrait, bright even lighting eliminating shadows, subtle fill from below, natural features preserved, minimal makeup aesthetic, clean white background —ar 3:4 —v 7 —sref 4478541 —sw 300”
High key means bright, shadowless lighting. This works great for beauty campaigns or beauty retouching practice. The lighting should feel like it came from a softbox directly in front and slightly above.
Prompt 5: Character Portrait
Prompt: “Character portrait showing lived-in face, evidence of life in skin texture and expression lines, eyes with depth and history, honest unposed moment, slightly asymmetric features —ar 3:4 —v 7 —sref 3312090 —sw 400”
Character portraits embrace imperfection. Use a higher style weight here because the aestheticweathered, honest, realis the point. Skip this prompt if you want commercial perfection.
Prompt 6: Low Key Dramatic
Prompt: “Low key portrait, single light source creating dramatic shadows, high contrast ratio, moody atmosphere, subject partially in shadow, cinematic film still quality —ar 3:4 —v 7 —sref 2296420 —sw 350”
Low key creates mystery. One light, heavy shadows, the subject half-visible. This is your noir look, your thriller protagonist, your moody editorial moment.
Prompt 7: Contemporary Editorial
Prompt: “Contemporary editorial portrait, unconventional pose or angle, fashion magazine aesthetic, artistic rather than commercial, slightly moody color grading —ar 3:4 —v 7 —sref 5561981 —sw 300”
Editorial pushes boundaries. This prompt works best when you want something that looks like it came from a high-fashion spread rather than a corporate headshot.
Prompt 8: Natural Light Window
Prompt: “Portrait lit by large window light only, soft directional quality without harshness, subject positioned for optimal light fall, subtle shadows defining features —ar 3:4 —v 7 —sref 2674195 —sw 350”
Window light is the classic portrait photographer’s tool. It produces flattering results that look natural because it behaves like actual available light. One large window, one subject, done.
Prompt 9: Street Portrait
Prompt: “Candid street portrait, available light only, unposed and unaware subject, urban environment providing context and light sources, documentary authenticity —ar 3:4 —v 7 —sref 4123956 —sw 400”
Street portraiture captures real moments with real people in real places. Use a higher style weight here because the aestheticraw, unpolished, authenticis non-negotiable.
Prompt 10: Silhouette Study
Prompt: “Portrait silhouette against bright background, distinct profile visible, rim lighting defining subject shape, high contrast, artistic interpretation —ar 3:4 —v 7 —sref 1234567 —sw 300”
Silhouettes are graphic and bold. The SREF code 1234567 is a placeholderuse any code that produces high-contrast black-and-white work. The subject becomes shape and shadow rather than detailed features.
“The difference between a good portrait and a great portrait lives in the lighting setup. SREF codes let you borrow proven lighting aesthetics without being a professional photographer.”
Anatomy of a Strong Portrait Prompt
A prompt that produces photorealistic portraits needs specific elements working together:
Subject description comes first. Age range, gender, ethnicity, facial structurebe specific. “Woman in her 40s with prominent cheekbones and laugh lines” beats “beautiful woman” every time.
Expression and mood come next. “Slight smile, eyes relaxed” beats “happy” because happy means different things to different people.
Wardrobe and styling establish context. A blazer signals professional. A worn denim jacket signals casual. Don’t leave this to chance.
Environment and background add storytelling. Plain backgrounds isolate the face. Contextual backgrounds tell you who this person is.
Lighting direction is non-negotiable for photorealism. Rembrandt, butterfly, split, looppick one and describe it. “Dramatic side lighting from the left” beats “good lighting” every time.
Camera and lens language helps V7 understand the technical finish. “85mm lens, f/1.4 bokeh” or “50mm documentary style” tells the model what photographic equipment to emulate.
Color grade sets the mood. “Neutral color grade” reads different from “warm golden tones” or “cool blue-gray palette.”
Parameters anchor everything. --ar 3:4 --v 7 --sref [code] --sw [weight]
Example structure:
photorealistic portrait of [detailed subject], [expression], [wardrobe], [setting], [lighting direction], [lens/camera], [color grade] --ar 3:4 --v 7 --sref [code] --sw 300SREF Weight Guide: Finding Your Sweet Spot
The —sw parameter range is 0 to 1000. Default is 100. Here’s how to pick:
| Use Case | —sw Value |
|---|---|
| Exploring style options | 100 |
| Balanced portrait work | 200-300 |
| Locked brand aesthetic | 400-500 |
| Maximum style transfer | 600+ |
Start at 300 for portraits. Generate. Compare.
If faces look too smooth or artificial, lower —sw to 200 and add “natural skin texture” to your prompt.
If the style reference isn’t coming through strong enough, raise —sw to 400.
If your subject looks distorted despite good style, drop —sw to 150 and simplify your subject description.
Common Problems and Fixes
Faces look like plastic. Add “natural skin texture, visible pores, freckles” to your prompt. Lower —sw by 50 points. Remove words like “flawless,” “perfect,” and “hyper-beautiful.”
The portrait looks too AI-generated. Strip out excessive descriptors. “Ultra detailed, hyperrealistic, 8k resolution” sounds impressive but actually pushes the model toward that over-processed AI look. Keep it simple.
The style reference dominates the subject. The subject should be recognizable even with heavy style transfer. If your subject gets lost, lower —sw or choose a reference with subtler texture.
Lighting is inconsistent. Pick a reference with one obvious light source. Multiple light setups in your reference confuse the style transfer.
The output feels generic. Your subject description is probably too vague. “Man” is generic. “58-year-old fisherman with weathered hands and crow’s feet from squinting into the sun” is specific.
SREF Code Sources and Discovery
According to SREF code libraries, there are over 4,800 style codes available for V7. You find codes three ways:
Style Explorer at Midjourney.com/explore. Browse by category, try styles instantly, save favorites.
Code libraries like srefcodes.com and midlibrary.io offer curated collections with preview images.
—sref random generates a unique code. Keep codes you like, discard ones you don’t.
Each code produces a different aesthetic. Photorealistic portrait codes typically live under “photographic,” “portrait,” and “cinematic” categories.
Ethical Considerations
AI portraits can look like real people. That comes with responsibility.
Never generate portraits of real individuals without consent. This includes celebrities, private individuals, and anyone who didn’t explicitly give permission.
Don’t use AI portraits for identity documents, fake testimonials, or deceptive marketing. These uses harm real people.
Document your workflow for commercial projects. Note which SREF codes you used, what references inspired your approach, and that the work is AI-generated.
AI portraits work well for fictional character studies, editorial moodboards, campaign concept art, lighting tests, and art direction references. They’re weaker for anything requiring actual photographic proof of a real person.
Quick Reference: 7-Step Portrait Workflow
- Write a specific subject description with age, ethnicity, expression, and styling
- Choose one lighting direction (Rembrandt, split, butterfly, etc.)
- Add camera and lens language for your desired finish
- Select a SREF code from a portrait photography category
- Set —sw to 300 as your baseline
- Generate and compare four variations
- Adjust —sw up or down based on results
This workflow gives you control without letting the style reference overpower your intent.
FAQ
What SREF codes work best for photorealistic portraits? Look for codes in the 1-10 million range that are tagged as “photographic,” “portrait,” or “cinematic.” Test multiple codes before committing. The best code depends on your specific subject and lighting setup.
Can I use my own photos as style references?
Yes. Upload your reference image to Midjourney, copy the URL, and use --sref [url] in your prompt. This creates a style transfer based on your specific image rather than a library code.
Why do my portraits look like plastic? The AI oversmooths skin. Fix it by adding “natural skin texture, visible pores” to your prompt, lowering —sw by 50 points, and removing perfection-describing words like “flawless” and “hyper-beautiful.”
What’s the difference between —sref and image prompts?
Image prompts tell Midjourney what to include in the scene. Style references tell Midjourney how to style what’s already in your prompt. You can use both together: --sref [url] for style, separate image URLs for content.
How do I maintain consistency across multiple portraits? Lock your SREF code and —sw value. Save your exact prompt structure. Generate a batch, pick your favorite, then iterate from that seed. V7’s prompt understanding is good enough that slight rewording keeps results consistent if your parameters stay fixed.
Do older SREF codes work in V7?
They work but may produce different results. V7 interprets style codes differently than V6. Add --sv 4 to force V7 to use the older style interpretation. For best results, find V7-specific codes rather than relying on legacy V6 collections.
Key Takeaways
- V7’s improved prompt accuracy makes photorealistic portraiture genuinely achievable
- SREF codes let you borrow proven photographic aesthetics without lighting expertise
- Start —sw at 300 and adjust based on whether you need more style or more subject control
- Specific subject descriptions beat vague descriptions every time
- Add “natural skin texture” to fight the AI plastic-skin tendency
- Test multiple SREF codes before committing to one for a project