Build 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.
Imagen 4 has no style-reference image input — its consistency comes from prompt fidelity. Write a brand prompt block once: "Studio photography style, soft window light from the left, [hex palette: warm beige #F5E6D3, deep moss green #4A5D3A, charcoal #2C2C2C], shallow depth of field, no text overlays." Save this block in a Text node on the canvas; every Imagen 4 node concatenates it with the per-asset subject prompt.
Imagen 4 reads hex codes literally. Write "warm beige #F5E6D3 backdrop" instead of "tan background" — the hex pin holds Pantone-grade color across 100+ generations. Pin three to five core brand colors in the saved block. Avoid color names alone; "navy" maps to a different color across runs.
On the canvas, wire the brand prompt block (Text node) into multiple Imagen 4 nodes — campaign hero (3:4), social square (1:1), Reels vertical (9:16), blog header (16:9), pitch deck cover (16:9). Each Imagen 4 node concatenates the brand block with its own subject prompt. The brand voice is identical across the five outputs because the prompt prefix is identical.
Imagen 4 ships in three tiers: Fast (drafts, lifestyle social), base (default, most assets), and Ultra (hero shots, product close-ups, anything that goes large on the homepage). Ultra adds finer texture rendering and tighter prompt adherence; Fast halves the cost for first-pass exploration. Pin the tier per asset class, not per generation.
Imagen 4's strongest property is literal prompt rendering. After every campaign batch, re-read the prompt and check the output against each named element — backdrop color, light direction, prop count, subject pose. If the prompt says "three ceramic vases on the left half of the frame," the output should show exactly that. Drift from explicit prompt elements is the failure mode to catch.
Save the entire canvas — brand prompt block, fan-out Imagen 4 nodes, parameter pins, color hex pins. Next quarter, swap subject prompts and re-run. The brand pipeline becomes reusable: same canvas, different campaign, same color and lighting voice. This is what compounds across a year of brand work.
Standing brand prompt prefix wired into every Imagen 4 node on the canvas. The hex codes are the load-bearing brand-lock command.
Studio photography style, soft window light from the left, warm beige #F5E6D3 backdrop, deep moss green #4A5D3A accent prop, charcoal #2C2C2C surface texture, shallow depth of field, single ceramic vase center frame, 3:4 aspect, magazine-quality crop, no text overlays
Social square that inherits the brand prefix. Subject prompt extends without overriding palette and lighting.
[brand prefix above] + Subject: lifestyle morning scene, hands lifting a ceramic mug from a wooden tray, soft golden window light, 1:1 social aspect
Blog header in the same brand world. The prefix locks studio voice; the subject prompt extends scope.
[brand prefix above] + Subject: blog header, wide composition of a ceramic studio interior, multiple vases drying on a wooden shelf, late afternoon raking light, atmospheric depth, 16:9 aspect
PPT cover with copy room baked into the composition. Pin Imagen 4 base tier here — Ultra is overkill for low-detail covers.
[brand prefix above] + Subject: deck cover, minimalist single vase silhouette against pale gradient, generous negative space top-right for headline copy, editorial restraint, 16:9 aspect
Lock brand colors as hex codes (#F5E6D3, not "warm beige") in the saved prompt block. Color names drift across runs; hex codes hold.
Pin the Imagen 4 tier per asset class. Hero = Ultra, default = base, draft/lifestyle = Fast. Switching tier mid-campaign breaks consistency.
Imagen 4 reads explicit composition language ("left half of the frame," "three ceramic vases"). Use literal counts and positions for prompt-adherence wins.
Avoid asking Imagen 4 to render headline text in-image. It is improving but Ideogram still wins for typography-driven brand work.
For lifestyle scenes with people, name the action explicitly ("hands lifting a mug from a wooden tray"). Generic "person using product" returns generic person.
Save the brand prompt block as a Text node on the canvas, then wire it into every Imagen 4 node. Editing the block once propagates to all assets.
Imagen 4 returns 1024x1024 or 2048x2048 outputs depending on tier, with strong prompt adherence and photographic realism. Hex-locked palette holds across 100+ generations from the same brand prompt block. Generation time: Fast 8-15s, base 15-30s, Ultra 30-60s. Output lands directly on the canvas as a single selected frame (no four-up grid). Route into ad-creative, social-graphic, or video starting-frame nodes downstream — the same anchor block ensures continuity across modalities.
Connect Imagen 4 with other AI models on Martini's infinite canvas. No GPU required — start free.
Get Started FreeMidjourney
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.
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