2.5Modes (V6, V7), style profiles
The essentials in 30 seconds
- Version = new render: obedience, realism, text — log --v, test before migrating.
- Simplify when migrating: inherited 'magic' keywords become useless — lighten the prompt.
- Show rather than describe: --sref (style), --cref (character) — key to the consistent series.
Running mini-case — Sonia locks her series before migrating
Final act: the café series pleases, but a new model version comes out. Sonia does not migrate mid-way: she finishes the series on the same version (noted --v), simplifies her prompts of old inherited keywords, and tests the new version on ONE reference prompt before deciding. Her library keeps for each visual: prompt, seed, version, --sref and parameters. Level 2 is delivered: format, steered randomness, image base, signature, controlled version.
Epilogue to thread 2.1-2.4: Sonia's poster becomes a traceable series, ready for level 3 (light, framing).
1. Model versions
Midjourney progresses by versions (V5, V6, V7…): better language understanding, realism, text in the image, anatomy. A prompt that worked on one version can give something else on the next — inherited 'magic' keywords often become useless, even harmful. Choose the version with --v (or in settings). It can be wise to stay on an earlier version to preserve the consistency of an ongoing series — like Sonia finishing the café menu before migrating.
2. References: from 'describing' to 'showing'
Major novelty: style personalization. Rather than describing a style with words (approximate), Sonia provides a reference image: for style (--sref), even for a character (--cref, section 3.5). The model also learns her preferences from her choices (personalization, style profiles). This is the move from 'describing' to 'showing' — key to the consistent series and brand identities (sections 3.5 and 5.1).
3. Migrating without breaking
Three situations justify staying on the old version: consistency of an ongoing series, aesthetic preference (a version's character may suit better), reproducibility (same prompt + same seed only give the same result on the same version). Sonia's method: systematically log the version in her library, retest her reference prompts on the new version before migrating, decide project by project. Versions and parameters change fast: check the up-to-date documentation.
🔎 Concrete example — 🔎 Worked example — Showing rather than describing
The café series, locked on one version and one signature:
café poster, steaming cup --sref [reference URL] --ar 4:5 --v 6
café menu, pastries --sref [same URL] --ar 4:5 --v 6
→ same version + same signature = consistent series.
(--cref to keep the same character, see 3.5)Reading: you convey a style from one image to another instead of redescribing it.
Lesson: log version and references to reproduce the series.
Section vocabulary
- Model version
- Midjourney model generation used (--v); each version markedly changes renders.
Simple example: Sonia finishes the café menu on the same logged version.
What it is not: a mere aesthetic setting with no effect on obedience.
Reflex: ongoing series = frozen version.
- Personalization
- Mechanism by which the model learns your aesthetic preferences from your choices.
Simple example: Sonia lastingly steers rendering toward her taste instead of the default.
What it is not: a different prompt at each generation.
Reflex: lasting taste = personalization.
- --sref / --cref
- --sref conveys a visual signature; --cref keeps the same character from one image to another.
Simple example: Sonia signs the menu in --sref, keeps the mascot in --cref.
What it is not: two interchangeable parameters.
Reflex: style = --sref; face = --cref.
- Series consistency
- Getting several images sharing the same style or character (see sections 3.5 and 5.1).
Simple example: Ten café visuals that look from the same hand.
What it is not: ten isolated random images.
Reflex: series = version + signature + traceability.
In practice — Hands on — Comparing versions and fixing a style
- Take a reference prompt and generate it on two different model versions: compare fidelity, realism and text rendering.
- Check whether your prompt holds keywords inherited from old versions that became useless; simplify it.
- Try a style reference (--sref) from an image whose aesthetics you like, on two different subjects.
- Log the version and references used in your prompt library, to reproduce the series later.
Ready-to-use script — justifying not migrating mid-series (30s)
'For the café menu, I don't switch versions mid-way: we would end with two aesthetics. I finish the series on the logged version, I simplify old keywords, and I test the new version on a single reference prompt. We migrate project by project — not on principle. Everything is logged: version, seed, --sref.'
Say while showing the homogeneous series then the isolated test on the new version.
Copy-paste template — series traceability sheet
SERIES: [client name] — frozen version --v 6
[subject A] --sref [URL] --ar 4:5 --v 6 --seed
[subject B] --sref [same URL] --ar 4:5 --v 6 --seed
[migration test] [reference prompt] --ar 4:5 --v 7 → compare before migrating
[character] [subject + --cref URL] → same face (see 3.5) New project
Fresh brief in Algiers: Sonia starts on the recent version and tests her reference prompts on it.
Ongoing series
Café menu half delivered: Sonia stays on the logged version until the last visual.
Inherited prompt
Old prompt loaded with keywords: Sonia simplifies it, the recent version obeys better.
Personal taste
Sonia enables personalization: the model learns her choices for her future series.
✓ Best practices
- Systematically log the version used (reproducibility).
- Simplify prompts: remove inherited keywords that became useless.
- Show style via --sref (--cref for a character) rather than describing it at length.
✗ Common pitfalls
- Switching versions mid-series (visual break).
- Keeping 'magic' keywords from old versions.
- Forgetting to document reference + version + parameters.
Key takeaways
- Each version (V5, V6, V7…) changes renders: better obedience, realism and text in the image.
- An old prompt can age badly: 'magic' keywords from old versions often become useless.
- Staying on an earlier version can be justified to preserve an ongoing series' consistency.
- References (--sref for style, --cref for a character) move from 'describing' to 'showing': key to the consistent series.
48h challenge — apply tomorrow
Take a reference prompt: generate it on 2 versions, simplify inherited keywords, try a --sref on 2 subjects, and log version + seed + references in your library.
- I can consciously choose --v and justify staying on a version.
- I can convey a style (--sref) and a character (--cref).
- I document version, seed and references to reproduce.
Frequently asked questions
Should I always use the latest model version?
For a NEW project yes by default (obedience, realism, text). But stay on the old one for: an ongoing series' consistency, an aesthetic preference, or reproducibility (same prompt + seed only exist on the same version). Log the version and retest your prompts before migrating.
How do I get several images that truly share the same style?
Three combined levels: describe style with a stable formula (level 3), freeze parameters (same version, stylize, seed), and above all show via --sref (style) or --cref (character, 3.5) — the approach of a visual AI identity (5.1). Document everything.
What do image references (--sref, --cref) allow?