Midjourney
Ship a quarter of brand-aligned visuals — campaign hero, social tiles, blog headers, deck imagery — in one Martini canvas using Midjourney v7. The model is the strongest pick for editorial-photography aesthetics and painterly brand worlds where mood and atmosphere carry the work. Lock a single style reference image, fan that anchor into every downstream node, and Midjourney holds palette, lighting, and tonal voice across hundreds of generations without redrafting the prompt each time.
Drop your canonical brand reference onto the canvas as an image node — a hero campaign shot, an editorial spread, or an existing brand asset. Wire it into Midjourney v7 as a style reference (sref) anchor. Every downstream Midjourney node on this canvas pulls from that anchor, which is how palette, lighting, and tonal voice survive across the quarter.
Midjourney v7 reads natural language. Skip "minimalist, beige, sunlit, modern" and write: "An editorial product still on a beige linen surface, soft late-afternoon window light raking across the frame, shallow depth of field, magazine-quality crop." Editorial sentences guide composition; keyword soup gives generic results.
On the canvas, duplicate the Midjourney node for each deliverable: campaign hero (3:2), social square (1:1), Reels vertical (9:16), blog header (16:9), pitch deck cover (16:9). Each duplicate inherits the same style reference. Change only the subject prompt — the anchor holds aesthetic continuity across all five outputs.
Midjourney v7 exposes --stylize (default 100, range 0-1000) and --weird (range 0-3000). Editorial campaign work lives at --stylize 100-250. Lifestyle social tiles can drift to --stylize 300-500 for more atmosphere. Reserve --weird for concept exploration only — it breaks brand consistency. Pin the values once per asset class, then leave them.
Brand drift compounds. Every 10-20 generations, drop the latest output next to the original style reference and check palette, lighting direction, and grain. If drift is visible, refresh the sref weight (--sw 100-200 is the working band) or re-run with the anchor in slot 1. Drift caught early is a re-prompt; drift caught late is a reshoot.
Once one campaign ships, save the entire canvas as a Martini template — the style reference, the duplicated Midjourney nodes, the prompt scaffolding. Next quarter, swap the subject prompts and re-run. The brand pipeline becomes reusable: same canvas, different campaign, same aesthetic.
Campaign hero anchor. The --sref + --sw combo is the load-bearing brand-lock command. Use this exact prompt structure for every hero across the quarter.
Editorial brand hero: a single ceramic vase on a beige linen surface, soft late-afternoon window light raking across the frame, shallow depth of field, magazine-quality crop, 3:2 aspect --sref [reference URL] --sw 150 --stylize 200
Social square that inherits the campaign world. The hands + lifestyle action expand the story without abandoning the aesthetic.
Same brand world, lifestyle insert: hands lifting a ceramic mug from the same beige surface, morning kitchen light, soft warm tones, shallow depth of field, 1:1 social --sref [reference URL] --sw 150 --stylize 250
Blog header keeps the studio context. Same anchor, same lighting language, wider crop. This is brand pipeline working as designed.
Blog header in the same brand world: wide composition of the ceramic studio interior, raking afternoon light, multiple vases drying on a wooden shelf, atmospheric depth, 16:9 --sref [reference URL] --sw 150 --stylize 200
PPT cover with copy room baked in. Lower --sw lets the negative space breathe; lower --stylize keeps the composition clean for headline overlay.
Deck cover: minimalist composition of a single ceramic vase silhouette against a pale gradient, generous negative space top-right for headline copy, editorial restraint, 16:9 --sref [reference URL] --sw 120 --stylize 150
Use --sref + --sw 100-200 as the standing brand-lock combo. Going above --sw 300 starts overcooking the reference; going below 80 lets drift creep in.
Pin --stylize per asset class, not per generation. Editorial 100-250, lifestyle 250-500, concept 500+. Keep it predictable.
Avoid --weird on brand work. It is for ideation only and will break the very consistency you are trying to lock.
Match aspect ratio to the placement (3:2 hero, 1:1 social, 9:16 vertical, 16:9 deck). Midjourney composes differently per ratio — wider crops invite more environmental context.
For text-in-image (campaign headlines), do not ask Midjourney — chain to Ideogram. Midjourney v7 still renders text inconsistently.
Save the working canvas as a template after the first successful campaign. The reference + node graph + parameter set become the brand pipeline.
Midjourney v7 returns four 1024-2048 wide variants per generation, configurable up to 2048x2048 on Standard tier. Sref-locked outputs hold palette and lighting at ~85-90% consistency across 50+ generations from the same anchor. Generation time runs 30-60 seconds per batch. Output drops onto the canvas as four selectable variants — pick the strongest, route the others to the reject pile, and feed the chosen frame into ad-creative or social-graphic nodes downstream.
Connect Midjourney v7 with other AI models on Martini's infinite canvas. No GPU required — start free.
Get Started FreeBuild a quarter of brand-aligned imagery — hero campaign frames, social tiles, blog headers, presentation imagery — on Martini using Imagen 4. Imagen 4 wins on photographic realism and on prompt adherence: it renders specific brand props, lighting setups, and color palettes with near-literal accuracy. Pair it with a saved style prompt block and a brand color reference, fan that into every downstream node, and the same canvas ships consistent campaign visuals across hundreds of generations.
View guideBlack Forest Labs
Generate a quarter of brand-aligned visuals on Martini using FLUX.2 — campaign hero, social tiles, blog headers, and deck imagery from one anchor canvas. FLUX.2 is the prompt-fidelity pick: it renders specific brand props, palettes, and compositions almost literally, without the embellishment Midjourney layers on. Pair it with a saved brand prompt block, fan that into every downstream node, and the model holds palette, lighting language, and compositional intent across hundreds of generations.
View guide