Storyboard frames with one look

Write prompts like a cinematographer and lock the visual style with --sref before animating frames.

Prompt like a camera crew

Frames that will later be animated in a video model need to read as shots from one film, not as separate artworks. Start every prompt with something like "cinematic film still" and then name what a cinematographer would set: shot type, camera or lens, depth of field, film stock, lighting and color grade. Keep the rest short. Long lists of adjectives make results harder to control, and AI filmmaking teachers advise cutting prompts back to the elements that matter, such as shot type and aspect ratio.

Pick the frame size first

Decide the aspect ratio before you generate anything: 16:9 for YouTube, 9:16 for Shorts, Reels and TikTok. Reframing later costs detail and often breaks the composition.

Lock the style

A style reference keeps color, texture and lighting the same across the sequence without copying objects or people. Use --sref with an image URL or a numeric style code. You can browse codes in the Style Explorer on the Explore page, or try --sref random. Once random has resolved to a code, reruns and variations keep it. Then paste that code into every prompt in the sequence. Adjust strength with --sw (0 to 1000, default 100). On V7 the weight has more effect with codes than with images.

When you use a style reference, describe what is in the picture, not how Midjourney should change the reference. "Rain-soaked alley at night, close-up of a courier" works better than "make this style but with a courier". Avoid style words that fight the reference.

Prepare for motion

Upscale inside Midjourney before you download, so the video tool starts from the sharpest frame. After animating, Curious Refuge recommends a separate upscale pass to win back detail that motion models tend to smooth away. Expect to reroll a lot: building a usable frame set usually takes several rounds of generations.