Prompt guide
A repeatable workflow for crafting prompts and refining the final look.
The same five moves, every time. A feeling first, then a structured prompt, then small variations, a finish, and a version someone else can reuse.
A useful prompt begins with a job, not decoration. Before writing, decide whether the image should sell a product, introduce a character, create a cinematic scene, explain an idea, or produce a social post. That choice keeps the prompt from turning into a long list of unrelated style words.
A strong first sentence usually answers four questions: who or what is the subject, where is it happening, what is the mood, and what should the viewer notice first.
A founder in a quiet studio reviewing product sketches under warm desk light.
That sentence gives the model a clearer direction than “cinematic professional portrait.”
AI prompts are easier to control when they are written in layers. Write the subject first, then the setting, then lighting, camera, texture, and output style. You can change one part without rewriting everything.
For portraits, keep identity and clothing in the subject layer. For product images, keep material, shape, and surface detail together.
For image to video, add motion as its own layer: camera movement, subject action, speed, and the final frame. Dance clips work the same way in Dance V2V: identity stays in the photo, and motion stays in the dance.
A reference image should tell the model what to preserve or reinterpret. If the face must stay similar, say that clearly. If the reference is only for mood, color, framing, or product shape, say that too. Mixing those intentions often creates inconsistent results.
When you recreate a prompt from the example gallery, use your own image if the goal is personalization. The public example is there to teach the composition and style. It should not force every new creation to copy the original subject.
Good prompt work is usually incremental. Change one variable at a time: ratio, lighting, camera distance, outfit, background, or model. If several things change at once, it is hard to tell why the result improved or failed.
Save the prompts that produce a usable direction, even when the image is not perfect. Those drafts become reusable building blocks for future scenes, campaigns, thumbnails, and video presets. Put anything you want the model to avoid in Undesired Content, so the next batch stays on the same path.
Before you use an output publicly, check for distorted anatomy, unreadable text, unintended logos, unsafe content, and misleading realism. For commercial work, confirm that you have the rights to any reference images, names, or brand assets that influenced the result.
The Imagesmith examples are chosen with these checks in mind. The best ones are not only attractive. They are understandable, reusable, and safe enough for other creators to learn from.