Preset Library
Styles and prompt templates
Every preset is a published prompt template. Below are the six styles, their intended use cases, and the exact string passed to the generation model.
Note: The thumbnails shown below are illustrative vector mockups, not actual model outputs.
Storytelling, essays, and documentary content
Cinematic
Prompt template
"Cinematic dramatic lighting, film color grade, rich shadows, shallow depth of field. Keep the subject of the reference image. YouTube thumbnail 16:9."
Visual traits: Deep shadows, film grain, dramatic directional lighting.
Tech reviews, commentary, and mainstream video topics
Bold and Vibrant
Prompt template
"High contrast, saturated colors, punchy, vibrant, YouTube thumbnail energy. Keep the subject of the reference image. YouTube thumbnail 16:9."
Visual traits: High color saturation, sharp outlines, bright highlights.
Tutorials, software demos, and productivity channels
Minimal Clean
Prompt template
"Clean minimal design, lots of negative space, soft lighting, elegant. Keep the subject of the reference image. YouTube thumbnail 16:9."
Visual traits: Generous negative space, studio softbox lighting.
Gaming, technology concepts, and animation
3D Render
Prompt template
"3D rendered aesthetic, clay/soft material, volumetric lighting, depth. Keep the subject of the reference image. YouTube thumbnail 16:9."
Visual traits: Volumetric light, isometric depth, smooth clay-like surfaces.
Gaming streams, anime reviews, and creative content
Anime and Manga
Prompt template
"Anime style, cel-shaded, vibrant colors, manga aesthetic. Keep the subject of the reference image. YouTube thumbnail 16:9."
Visual traits: Cel-shaded colors, dynamic line work.
Podcasts, music, and night-themed vlogs
Neon Glow
Prompt template
"Neon glow, cyberpunk aesthetic, vibrant magenta/cyan highlights, dark background. Keep the subject of the reference image. YouTube thumbnail 16:9."
Visual traits: Dual-tone magenta and cyan accents against a dark backdrop.
Generation uses google/gemini-3.1-flash-image via OpenRouter by default. The model endpoint accepts reference images via input conditioning and renders directly to 16:9.