Midjourney Draft + Random SREF: Find a Style in One Pass

Why Style Hunting Needs a Separate Pass

Most creators burn Fast time guessing a look: they reroll the same prompt, tweak adjectives, and still land on Midjourney’s house aesthetic. Style hunting is different from composition hunting. You already know the subject—you need a visual dialect that can survive a whole campaign.

Midjourney’s Draft Mode already gives you a 24-image cheap grid. Pairing it with --sref random turns that grid into a style sampler: each draft tile can pull a different style reference, so one job explores aesthetic space instead of repeating the same vibe twenty-four times.

If you only need the Draft → Vary cost loop, read our Draft Mode workflow. This tutorial assumes Draft is on and focuses on random SREF → lock → reuse.

Step 1 — Freeze the Subject Sentence

Write one clear subject line and do not change it during the hunt. Style noise is useless if the subject also drifts.

Example (keep this string fixed):

editorial portrait of a ceramicist at a sunlit workbench, soft window light, shallow depth of field

Rules for the frozen sentence:

1. Name the subject and place.
2. Add one lighting cue.
3. Skip style adjectives like “cinematic masterpiece” or “trending on ArtStation”.
4. Put parameters at the end only.

Step 2 — Run Draft With Random Style References

On the web Create page, enable Draft Mode (or append --draft), then add --sref random:

editorial portrait of a ceramicist at a sunlit workbench, soft window light, shallow depth of field --ar 4:5 --draft --sref random

What you should expect:

• A large low-res batch (Draft’s 24-tile exploration lane on current V8.x web Draft).
• Noticeably different palettes, brush feels, and photographic treatments across tiles.
• None of the tiles are finals—treat them as direction chips.

Tip: run two hunts with the same subject if the first grid is too narrow. Still change only the random style pass, not the subject sentence.

Step 3 — Score Tiles Like a Creative Director

Do not pick “the prettiest face.” Score for series survival:

Mark 2–3 keepers max. More than three usually means you have not decided.

Step 4 — Promote Keepers, Then Harvest the Code

For each keeper:

1. Click Vary (or your Studio equivalent) to regenerate at full SD/HD quality.
2. Open job details and record the resolved style reference code Midjourney attached.
3. Store the code in a project note: name, subject, date, and intended use (packaging, social, hero).

Re-test the code outside Draft:

editorial portrait of a ceramicist at a sunlit workbench, soft window light, shallow depth of field --ar 4:5 --sref <winning-code> --sw 100

If the look collapses, try --sw between 60 and 140 before abandoning the code. Extreme weights often fight the subject.

Step 5 — Build a Tiny Style Deck (Not a Museum)

A usable deck is three codes, not thirty:

1. Hero look — primary campaign voice.
2. Support look — softer or higher contrast sibling for secondary assets.
3. Wildcard — for mood experiments; keep it out of client decks until proven.

For each code, generate three different subjects with the same --sref and identical --sw. If identity drifts wildly, the code is decorative, not operational—retire it.

Step 6 — Combine With Personalization Carefully

--p (Personalization) and a locked --sref both push taste. Stacking them without a plan creates muddy hybrids.

Practical rule used in MidassAI Studio projects:

• Style hunt with --sref random without --p.
• After you lock a code, A/B one frame with and without your default --p.
• Keep the stack only if the difference is intentional and repeatable.

Common Failure Modes

• Changing the subject mid-hunt — you cannot tell style from content.
• Promoting every cute tile — burns HD budget and leaves no brand spine.
• Never recording the resolved code — luck cannot be scheduled for next week’s assets.
• Using Draft tiles as finals — Draft is exploration resolution, not delivery.

Studio Checklist

1. Save the frozen subject prompt as a Studio preset named style-hunt-subject.
2. Run Draft + --sref random once (or twice).
3. Vary ≤3 keepers to HD only when the campaign needs finals.
4. Log winning codes in the project brief.
5. Produce the series with fixed --sref / --sw, changing only scene text.

Try Midjourney in MidassAI Studio to run this hunt loop with saved presets and a reusable style deck.

Related articles

• Conversational Draft Mode: Iterate Prompts by Talking
• Midjourney Draft Mode: Explore 24 Options Before You Spend…
• Multi-Image Prompts in Edit Model: Up to Four References
• Prompt Like a Photographer: Natural Language for V8
• Midjourney V8 SD vs HD: When to Pay for Native 2K