Midjourney v8 is a significant step up from v6 in coherence, photorealism, and prompt adherence — but it rewards a different prompting style than earlier versions. If you are still writing 40-word keyword chains, you are getting worse results than someone writing 12 sharp words. Here is how v8 actually works and how to get professional output from it in 2026.
What changed in 2026
- v8 launched with substantially improved photorealism, especially for human faces, hands, and scene lighting. The "uncanny valley" issues that plagued earlier versions are largely resolved at default settings.
- Midjourney added a web editor with inpainting, outpainting, and layer controls — meaning you no longer have to do all composition work in Discord or the app alone.
- Character Reference (--cref) became reliable for consistent characters across images, opening up visual storytelling and brand mascot work.
- Commercial licensing is clearer in 2026 — all paid tiers ($10–$120/month) include commercial use rights; free tier generations are not commercially licensed.
Understanding v8 defaults
v8 applies an "aesthetic layer" that makes outputs look polished by default. This is great for general users but can conflict with your vision if you want a raw or specific look.
| Behavior |
Default (v8) |
With --style raw |
| Aesthetic interpretation |
Heavy beautification |
Prompt literal |
| Lighting |
Cinematic, dramatic |
Neutral, as specified |
| Colors |
Boosted/stylized |
Natural/specified |
| Good for |
Social graphics, hero images |
Product shots, technical illustration |
Start with --style raw when your prompt specifies a precise look. Use defaults for mood pieces where "beautiful" is the goal.
Core parameters to know
| Parameter |
What it does |
Common value |
| --ar |
Aspect ratio |
16:9, 4:3, 1:1, 9:16 |
| --q |
Quality (render time/detail) |
1 (fast) or 2 (detailed) |
| --style raw |
Disables aesthetic layer |
Toggle on/off |
| --cref [URL] |
Character reference |
URL to your character |
| --sref [URL] |
Style reference |
URL to style source |
| --cw |
Character weight (0–100) |
50–75 for likeness |
| --chaos |
Variation level (0–100) |
20–40 for consistency |
| --no |
Negative prompt |
--no text, watermarks |
How to write prompts for v8
v8 is trained on intent, not keyword density. A well-structured sentence outperforms a comma-separated list of adjectives.
Weak prompt: beautiful woman, golden hour, cinematic, bokeh, editorial, fashion, luxury, 8k, photorealistic, stunning
Strong prompt: Editorial fashion portrait, model in tailored camel coat, golden afternoon sun through warehouse window, shallow depth of field --ar 3:4 --style raw --q 2
Prompt structure that works: [Subject + key detail] + [setting/environment] + [lighting] + [mood/treatment] + [parameters]
Include what makes the shot specific. Skip generic quality adjectives — v8 already renders at high quality.
Character Reference workflow
- Generate (or find) a reference image of your character.
- Upload it to Midjourney and copy the attachment URL.
- Add
--cref [URL] to any subsequent prompt.
- Use
--cw 50 for loose style similarity, --cw 100 for tight likeness.
This is the primary way to maintain consistency across a project. Works best with faces and distinctive designs; less reliable for complex poses.
Style Reference workflow
--sref pulls color palette, texture, and composition style from a reference image without copying the subject.
A mountain landscape at dusk --sref [URL-to-painting] --ar 16:9 --q 2
You can combine multiple sref URLs with weights: --sref URL1::2 URL2::1
How to pick the right approach
- Quick concept or mood board? Use defaults, --chaos 30, vary aspect ratios.
- Brand asset / commercial deliverable? Use --style raw, --q 2, explicit lighting.
- Consistent character across scenes? Set up --cref workflow first, test with --cw 50 and 75.
- Match an existing visual identity? Use --sref on approved brand imagery.
- Illustration rather than photo? Add "vector illustration," "flat design," or "ink drawing" to shift rendering mode.
Common mistakes
Using v6 prompt style in v8. Keyword dumps made sense when models needed cues; v8 reads sentences. Restructure your prompts.
Ignoring aspect ratio. Default is 1:1. If you are making hero images, social posts, or video thumbnails, set --ar explicitly or you are cropping every output.
Not using --no for recurring problems. If you keep getting text artifacts or extra limbs, add --no text, extra fingers rather than hoping the next generation fixes it.
Assuming one generation is the output. Use Vary (Subtle) and Vary (Strong) on promising results. The first grid is a draft.
Skipping upscale for deliverables. The default resolution is fine for comps. For print or large-format digital, use the Upscale (Creative) option.
What to skip
- Writing 60-word prompts. More words past ~20 dilute intent in v8; trim ruthlessly.
- The free tier for commercial work. It does not include commercial rights; use a paid plan.
- Generating without a brief. Vague prompts waste credits. Know the aspect ratio, treatment, and required elements before you start.
FAQ
Is Midjourney v8 better than DALL-E or Flux for photorealism?
In 2026, all three are competitive. Midjourney v8 leads on aesthetic coherence and faces; Flux excels on prompt adherence for complex scenes. See How to use Flux image in 2026 for a direct comparison workflow.
Can I use Midjourney v8 outputs commercially?
Yes, on any paid plan. Free tier outputs are not commercially licensed. Check the Midjourney terms for your specific use case (advertising, merchandise, etc.).
How do I get consistent characters across images?
Use --cref with a clean reference image at --cw 75. Expect some variation — full character lock requires iteration and often prompt refinement.
What resolution do Midjourney v8 outputs come at?
Default generations are ~1024px on the long edge. Upscaling options reach ~2048px. For print above A4, use the Creative Upscale or pass through an AI upscaler like Topaz.
Where to go next
See How to use Flux image in 2026, How to use Firefly in 2026, and AI for graphic designers in 2026.