Verified against Midjourney · 2026-07-29
Hold a whole icon set visually consistent with --seed, no reference image needed
A text-only consistency workflow using a locked --seed value across many independently-run icon prompts, for cases where a reference-image-based approach is overkill — a coherent set built purely through disciplined, repeated wording and one shared seed.
The prompt
Ready to copy — highlighted parts are example details you can swap.
SET CONCEPT a set of kitchen and cooking icons for a recipe app — the overall theme uniting every icon in this set, restated identically in every individual icon prompt. STYLE TEMPLATE a minimalist flat-icon design, thin 2px outline, single accent color of warm terracotta, no gradients, no shadow, centered on a plain white background — the exact rendering style (line weight, color treatment, level of detail) written once here and then copied verbatim into every icon's prompt with no rewording between icons, since even small unintentional wording drift between otherwise-similar prompts is often enough to shift the rendered style noticeably across a set. FIRST ICON IN THE SET a whisk — generate this one first, without a seed specified, and review it against the style template before locking anything; committing to a seed from a result that doesn't actually match the intended style just propagates that mismatch through the entire rest of the set. SEED TO LOCK Once the first icon's style is approved, note its seed number and reuse that exact value — 2847193056 — as --seed on every subsequent icon in the set. REMAINING ICONS a rolling pin, a coffee cup, a mixing bowl, a chef's knife — generate one at a time, each using the identical style template text and the identical locked seed, changing only the specific subject named for that icon. WHAT SEED-LOCKING ACTUALLY GUARANTEES A locked seed biases the model's starting point toward a similar overall composition and rendering approach across prompts that are otherwise worded identically — it narrows random variation, it does not force pixel-perfect consistency the way a reference image would. Some acceptable variation in exact line placement or minor proportion between icons is normal and expected; if the variation crosses into visibly different styles (a different color treatment, a different level of detail) across two icons that used the identical seed and style template text, that is a sign the wording drifted between the two prompts somewhere, not a seed failure. WHAT TO AVOID --no text, watermark, background texture, drop shadow PARAMETERS PER ICON a minimalist flat-icon design, thin 2px outline, single accent color of warm terracotta, no gradients, no shadow, centered on a plain white background, depicting a rolling pin --no text, watermark, background texture, drop shadow --seed 2847193056 --ar 1:1 --v 7 OUTPUT One icon per prompt run, sharing a seed-biased overall style and composition approach across the full set, differing only in the specific object depicted.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
--seed fixes the initial noise pattern the generation process starts from, which biases otherwise-identical prompts toward a similar overall composition and rendering approach rather than each one starting from an unrelated random starting point — this is a real, useful lever for icon-set consistency precisely because it requires no reference image at all, which matters for a workflow where the icons don't share a face, an object, or a color palette in the --sref or --oref sense, only a shared rendering philosophy that text description alone has to carry. Generating the first icon without a locked seed and reviewing it before committing to anything addresses a real ordering trap: locking a seed from a result that doesn't actually match the intended style template just propagates that same mismatch through every remaining icon in the set, since the seed is reinforcing whatever composition and rendering tendency that first result happened to land on — a bad first icon, seed-locked, produces a consistently bad set rather than a consistently good one. Copying the style-template wording verbatim into every icon's prompt, rather than paraphrasing the same intent slightly differently each time, matters because seed-locking only narrows variation between prompts that are otherwise close to identical — it does not correct for or override wording that has actually drifted. Two prompts that describe the same intended style in slightly different words, even with the same seed, can still land on visibly different renderings, because the seed is biasing the starting point of two meaningfully different text-conditioned generations, not forcing two different descriptions to converge on one result. The explicit distinction between what seed-locking guarantees (a narrowed range of variation) and what it does not (pixel-perfect matching) sets a realistic, checkable expectation that prevents a common frustration: a user expecting a locked seed to behave like a reference image — forcing near-identical results — will read normal, acceptable minor variation between icons as a failure of the technique, when it's actually working within its real, more modest scope. Distinguishing "some line-placement variation" (expected) from "a genuinely different style or detail level" (a real sign of wording drift) gives a concrete standard for telling the two apart rather than a vague sense that something in the set "feels off."
Verified against
Midjourney v7 · 2026-07-29
Changelog
- 2026-07-29 — Initial publish, verified against Midjourney v7 --seed for text-only icon-set consistency.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
