Store banners, launch posters, and community-event art fail for the same reason on most image models: the prompt asks for readable slogans. Nanomelon is strong at atmosphere, props, and lighting. It is weak at crisp multi-language type. This guide shows a poster workflow that keeps credits useful and keeps your brand text under your control. Pair it with prompt craft, ecommerce sets, and credits.
Before you open the homepage generator, write five lines on paper: subject (person, product, or empty stage), camera distance (wide street, mid character, tight product), light direction, color mood in concrete materials (wet asphalt, paper lantern glow, matte paint—not “cinematic”), and negative space for type. Negative space is the commercial difference. If every corner is filled with neon and smoke, you will have nowhere to place a title without covering faces.
Aspect ratio is part of the brief. Use 3:4 or 4:5 for vertical event posters and phone stories. Use 16:9 for storefront screens and web heroes. Use 1:1 only when the deliverable is truly square. Generating 16:9 and cropping to 9:16 later throws away the composition the model balanced for you.
Start in text-to-image when you do not yet have a hero photo. Name props instead of vibes. Bad: “epic summer festival poster, viral, masterpiece.” Better: “night market alley, paper lanterns with warm internal glow, wooden stall tables, steam from food trays, empty wall on the upper third for a title, illustration, soft outlines, no logos.” The empty wall clause is intentional. Models often ignore “leave space,” but mentioning a flat wall or sky band raises the odds.
Keep brand names, trademarked characters, and celebrity likenesses out of the prompt. Describe a generic performer or an anonymous vendor. Policy blocks waste time even when credits are refunded; successful near-misses that look like a known IP still create legal risk for you under our Terms.
If the poster must feature a real SKU or a consistent mascot, shoot or export a clean plate first. Upload it in image-to-image and describe only the environment change: “keep bottle and label identical, place on concrete ledge at dusk, city lights soft bokeh behind, leave dark sky for title.” Restating the whole product as if inventing it from text invites label melt. When type on packaging warps, plan to composite the real label later—that is normal for catalog work and for posters.
Quantity stays at 1 until the plate works. Raising quantity multiplies cost on an unfinished brief. Failures and policy refusals do not deduct credits; pretty-but-wrong successes do.
Composite the title, date, venue, and CTA in Figma, Canva, Photoshop, or Affinity. Choose contrast against the plate you generated: light type on a dark sky band, or a solid bar if the AI scene is busy. Do not ask Nanomelon to “write Summer Sale 50% Off in clear English across the top.” You will get alphabet soup and spend credits fixing it.
If you need a small hand-lettered look for atmosphere only (illegible chalk menu in the far background), say “blurred chalk board, unreadable marks.” That reduces the chance the model invents fake brand words in the midground.
If Pass A fails three times, the brief is wrong—usually too many adjectives and too few objects. Rewrite the object list before changing models. Model hopping before the brief is stable wastes the starter stack.
Leave public visibility off for unreleased campaigns. Public Images require staff approval and expose prompts. Competitors can read your brief. For learning, study long object-list prompts such as gallery 105; for client files, keep the account private.
Do not generate medical before/after claims, guaranteed investment returns, or deepfake endorsements of real people. Those areas collide with ad policies and publisher rules whether or not AdSense is live. Stick to atmosphere, product context, and event mood.
On the homepage, write the five-line brief first, set quantity to 1, then Generate. Composite slogans offline.