Midjourney for UX Designers: Fix Style Drift in UI Concept Images

You are a senior visual design systems specialist with 11 years of experience creating AI-generated UI concept imagery for SaaS product teams where visual consistency across a multi-screen prototype presentation determines whether a design direction is approved or sent back for revision. Help me write a texture generation prompt so I can reduce post-processing time and produce a cohesive set of UI concept images that maintain identical visual style, color temperature, and interface density across every screen in the presentation. My situation: - Product type and design phase: [e.g., "a B2B project analytics dashboard in early-stage concept exploration — presenting three screen concepts to the VP of Product for directional sign-off before wireframing begins"] - Target visual style: [e.g., "a clean, high-contrast dark-mode UI with cool blue accent colors, fine grid lines, and data visualization elements — referencing Linear and Vercel's design language without directly copying either"] - Current style drift pattern: [e.g., "the first image generates the correct dark-mode palette and accent color, but by the third image the background has shifted from near-black to dark grey and the accent color has warmed from blue to teal"] - Image dimensions and output format: [e.g., "1920x1080 landscape for presentation slides — images need to function as full-bleed slide backgrounds with UI elements occupying the center 60% of the frame"] - Generation tool and current parameter setup: [e.g., "Midjourney v6 — currently using --ar 16:9 and --style raw but no seed locking or style reference image, which is likely causing the drift"] - Brand color constraints: [e.g., "the company brand palette is #0A0F1E (background), #1E2D4A (surface), #3B82F6 (primary accent), and #E2E8F0 (primary text) — all generated images must stay within this palette"] - Presentation context: [e.g., "12 concept images across three screen types — dashboard overview, detail panel, and empty state — four images per screen type, all must read as belonging to the same product"] Deliver: 1. A master seed-locked prompt template for the dark-mode dashboard style — a single Midjourney prompt that establishes the background color, surface color, accent color, grid density, and UI element type with a fixed --seed value and a --sref or --cref parameter strategy for locking visual continuity across all 12 images 2. Three screen-type prompt variants built on the master template — one for the dashboard overview (data-dense, multiple chart types, navigation rail visible), one for the detail panel (single data subject, expanded view, sidebar context), and one for the empty state (minimal UI, single onboarding illustration element, centered layout) — each variant changing only the layout description while preserving every style parameter 3. A color enforcement insert — a precise color language block that translates the four brand hex values into Midjourney-readable descriptors (e.g., "deep navy #0A0F1E background, steel blue #1E2D4A surface panels, electric blue #3B82F6 accent highlights, off-white #E2E8F0 text elements") to be appended to every prompt in the set 4. A negative prompt block — a list of ten exclusion terms that prevent Midjourney from introducing warm tones, light-mode elements, photographic texture, gradient backgrounds, or decorative illustration styles that conflict with the target UI aesthetic 5. A generation sequence protocol — a step-by-step process for generating all 12 images in a single session using seed locking and style reference chaining, covering the order in which to generate (master first, screen variants second, empty state last), how to extract the seed from the first successful generation, and how to apply it to all subsequent prompts 6. A post-generation consistency audit checklist — eight visual criteria to check across all 12 images before presenting (background hex match, accent color match, font weight consistency, grid line visibility, UI element scale consistency, lighting direction, depth of field uniformity, and aspect ratio compliance) with a pass or fail threshold for each 7. A prompt iteration log template — a structured record for each generation attempt covering the prompt text used, the seed value, the output rating on a 1-5 scale against each of the eight audit criteria, and the specific parameter change made for the next iteration, building a reusable generation reference for future concept presentations **Treat every prompt parameter as a style contract — write the master template and all variants assuming that any unspecified visual element will be interpreted inconsistently by the model, and specify every visual dimension explicitly rather than relying on the model to infer from context.**