Image models read prompts in one of three ways. Most of them, Flux, Nano Banana, GPT Image, Firefly and Midjourney v7, read plain sentences. Midjourney also takes short fragments with flags, and Stable Diffusion wants tags. Character Prompt Builder writes the same character in the form you pick.
Pick a prompt style
On the result step, under Prompt style, there are three choices:
- Sentences, the default: natural-language prose with no flags. Right for Flux, Nano Banana, GPT Image, Midjourney v7, Firefly, Veo and Sora.
- Midjourney: comma fragments, with
--arfor the aspect ratio and--style raw --v 7at the end. For a different version or Niji, change the last flag by hand. - Stable Diffusion: lowercase tags for SDXL and SD 1.5, with quality tags added (masterpiece, best quality, highly detailed). For SD 1.5, cut the quality tags if the prompt runs long.
A line under the choice says what the style does. The builder remembers your pick for next time.
Image or video
The result has an Image prompt tab and a Video prompt tab. The video prompt describes the character as footage: what they do, how the camera moves, the pacing, the clip length and the aspect ratio. It uses what you answered in the Motion step, so fill that in before you copy a video prompt.
For Veo or Sora, keep Sentences and use the Video tab. Veo also makes sound, so fill in Ambient sound in the Motion step, such as "rain on a tin roof, distant traffic".
Copy all four
Copy all four copies the image prompt, the video prompt, the identity block and the negative prompt together, each with a heading. That is handy when you keep your prompts in a document, or when you will use the same character in an image tool and a video tool.
Which style for a new model
Start with Sentences: nearly every model made since 2024 reads prose well. Switch to Stable Diffusion only for models built on SD, such as most anime checkpoints. The character's words stay the same; only the form around them changes.