Model / tool keywords (people search these)
- “text to image”
- “image to image”
- “inpainting” / “outpainting”
- “reference image” / “style reference”
- “control image” / “pose control” / “depth control”
- “LoRA” / “fine-tune”
Visual style keywords
- photorealistic / cinematic / editorial / product photography
- flat illustration / vector / isometric / 3D render
- pixel art / anime / manga / watercolor / oil painting
- brutalist / neo-brutalism / Swiss typography / Bauhaus
Composition + lighting
- rule of thirds / centered / negative space
- softbox lighting / rim light / golden hour / studio backdrop
- shallow depth of field / bokeh / macro
Typography-in-image (hard mode)
- “clean headline” / “legible text” / “no gibberish”
- “poster design” / “packaging mockup” / “brand banner”
FAQ
Q: Why do some models struggle with text?
A: Because the model learns text as pixels, not as characters. You’ll usually get better results by adding text later in a canvas editor.
Q: What are the 3 prompt parts that matter most?
A: Subject, composition, and style. Everything else is a modifier.
Q: Should I use negative prompts?
A: If your model supports it, yes—use them to remove artifacts (extra fingers, warped logos, messy background).