Realism
Realism responds well to photographic language — lighting setups, lens characteristics, and real-world setting descriptions produce the most accurate output. Avoid illustration or painting style tags entirely; they push the model toward an uncanny middle ground rather than blending styles. Add skin detail terms explicitly:detailed skin texture, natural pores, soft subsurface scattering improve close-up portrait quality significantly. For anatomy accuracy, pair with the Better Breast or BetterCock LoRA.
Avoid: anime, illustrated, painted, cartoon, 3D render — these degrade output quality on Realism without producing a hybrid style.
Anime
Anime responds to character-description language more than scene-description language. Hair color, eye style, outfit details, and emotional expression drive the output more than lighting or camera terms. Quality tags that work well:detailed linework, soft shading, clean lines, high detail, 2D illustration. Avoid photorealistic quality tags like photorealistic, natural skin, subsurface scattering — they produce muddy output.
For explicit content, Anime requires more explicit prompt terms than Realism to produce the same result — be more direct about what you want.
Cinematic
Cinematic is driven by lighting and color grading language. Without strong lighting terms the model defaults to a flat editorial look. Push it with:dramatic side light, single overhead light, neon fill light, golden hour, deep shadows, high contrast.
Color grading terms work well: teal and orange grade, desaturated cold tones, warm filmic grade. Treat the prompt like a film brief — setting and mood matter as much as subject.
Avoid: flat lighting terms, studio white background, anime or illustration style tags.
3D
3D produces the best results with terms that describe rendered surfaces:smooth skin, volumetric lighting, soft shadows, subsurface scattering, high poly, CGI render. Vague style terms produce inconsistent output — be specific about the render quality you want.
Character design language works well: stylized proportions, large eyes, detailed outfit, game-art style. Avoid photorealistic skin terms — they conflict with the model’s rendering aesthetic.
Cartoon
Cartoon responds to graphic design language more than photographic or painterly language. Describe the illustration style directly:retro pin-up, comic book style, bold outlines, flat color fill, 1950s illustration, graphic poster.
Avoid: photorealistic, cinematic, subsurface scattering, depth of field — these terms produce no effect or degrade the flat graphic quality that makes Cartoon useful.
For explicit content, describe anatomy in flat graphic terms rather than realistic ones.
Versatile
Versatile is the least opinionated model — it responds to a wider range of style combinations without breaking. Use it when mixing style tags that would conflict on a specialized model. It performs best when the prompt is specific about subject and setting and loose about style. Heavy style direction works better on specialized models.Prompting Basics
General prompt structure and element order.
Advanced Prompting
Weights, negatives, and token order.