Nanomelon exposes several upstream engines. Some lean illustration; some lean camera. If your prompt says “anime girl, 85mm photo, ultra realistic skin,” you are asking for two defaults at once. The result is usually a beauty-filtered hybrid that satisfies neither a character sheet nor a client photo brief. This page separates the two paths and shows which clauses belong to which.
Lead with the finish:
Then add subject and place. If the medium is buried at the end after twenty mood words, many models ignore it. One medium name is enough; repeating “masterpiece anime best quality” does not buy cleaner lines.
Useful: line weight (thin clean outline vs brush ink), shading model (cel / soft / hatching), eye style (large stylized vs normal proportion), and print tradition if relevant (new-year print, manga panel, healing comic). The festival example on gallery 105 works partly because it commits to “whimsical healing comic” and then lists objects—lantern horse, fireworks, architecture—instead of only saying “Chinese style.”
Less useful: “anime” alone, “waifu,” “trending on pixiv.” Those invite the most common training average. If eyes go too large, say “normal eye size for a teen character, soft highlight.” If the horse lantern turns into a plush toy, repeat “paper sculpture lantern.”
Useful: lens hint (35mm environment, 85mm portrait), light direction, white-balance note, and surface reality (“visible pores,” “brushed metal,” “condensation on glass”). Name the place with props. Avoid stacking “8k cinematic volumetric” unless you want haze.
Less useful: “realistic” alone. Every photo model already aims there. If skin looks plastic, add “no beauty blur.” If backgrounds melt, simplify the scene instead of adding “sharp focus.”
Keep the same subject sentence. Run quantity 1 twice:
Do not change wardrobe between A and B on the first test. You are measuring medium adherence, not fashion. Use one model for both runs; switching engines mid-test mixes two default stacks. Credits and model habits: models and credits.
Legitimate hybrids exist: “photo of a physical paper diorama,” “3D render with subsurface skin.” Accidental hybrids happen when you paste a photo lens stack onto an anime character brief. If you need a character that later appears in a photo-like set, generate the character sheet in illustration mode first, then use image-to-image with a careful change prompt—or composite in an editor. Expecting one click to invent a consistent IP across styles is how you burn a balance.
Do not aim for a trademarked mascot or a living celebrity’s face. Write original characters. Public gallery works show prompts in clear text; do not publish private photos of people who did not agree. About our stance on misleading synthetic media: About.
Run the A/B medium test on the homepage, then read the object-list breakdown on gallery 105.