Reference
This AI video prompt generator workflow turns a still reference into motion direction. RefPrompt separates what the scene looks like from what should happen over time: camera path, subject action, pacing, environmental movement, continuity rules, and likely failure modes.
RefPrompt turns a reference frame into video-ready direction: camera path, scene duration, motion hierarchy, environmental movement, continuity rules, and negative constraints that protect the shot.
What this workflow covers
Follow the reference analysis through reusable image and video prompts, a Image-to-video adaptation, a real result, practical limitations, and focused negative constraints.
Try it here
Upload a reference below and RefPrompt returns a copy-ready video / motion prompt — no sign-in, right on this page.
Style Breakdown
The style breakdown translates static visual cues into motion prompts: what should move, what should stay stable, and which scene details anchor the clip.
Image Prompt
The still prompt creates a strong keyframe before animation.
Rain-slick underground transit tunnel, centered vanishing point, repeating concrete ribs, reflective wet floor, cool white overhead strip lights, faint amber exit glow, low floor mist, cinematic infrastructure keyframe, stable perspective, no people, no vehicles, no signage.Video Prompt
The video version adds camera movement, timing, and continuity.
Create a 6-second AI video shot. Start from a locked centered view down a wet concrete transit tunnel and move forward in one slow stabilized dolly. Let floor reflections and low mist drift subtly as the camera advances. Keep tunnel geometry, vanishing point, concrete ribs, lighting, and rails stable. No cuts, no handheld shake, no fast zoom, no changing architecture.Midjourney Version
Midjourney can create the controlled keyframe before video generation.
rain slick underground transit tunnel, centered vanishing point, repeating concrete ribs, reflective wet floor, cool white strip lights, faint amber exit glow, low mist, cinematic keyframe --ar 16:9 --style raw --v 6Image-to-video Version
Use this version for video tools that start from a generated keyframe.
Use the reference frame as the first frame. Animate a single slow forward camera glide for 6 seconds. Preserve the tunnel shape, concrete ribs, vanishing point, strip lights, wet floor, rails, mist, and reflection behavior. Avoid cuts, wobble, flicker, fast zooms, changing architecture, or new objects entering frame.Result & Limits
The frame has a strong vanishing point, repeating concrete ribs, reflective floor, and visible atmosphere. Those cues make it ideal for a prompt that describes motion through space instead of just a static scene.
Prompt used in this case
The video prompt asks for a six-second stabilized dolly forward, subtle floor reflection movement, unchanged tunnel geometry, and no cuts or wobble.
Result read
The generated keyframe preserves tunnel depth and reflection logic. The main risk in video is geometry drift, so the prompt repeats stable vanishing point, unchanged concrete ribs, and controlled movement.
Negative Prompt
Video prompts need constraints for temporal stability.
warped tunnel, bending concrete ribs, flickering reflections, unstable vanishing point, jitter, handheld shake, jump cuts, fast zoom, rolling shutter, changing rails, noisy shadows, overexposed lights, fisheye distortion, new objects, low-resolution wet floor texture.CTA
Use the same page structure inside RefPrompt: start from a visual reference, then copy the prompt version that fits the tool and channel.
Try this workflow with RefPrompt.It should include the visual reference, shot length, camera movement, subject or scene motion, pacing, continuity rules, and negative constraints. RefPrompt structures those details from the source image or video.
An image prompt describes one frame. A video prompt also describes what changes over time: camera path, subject action, environmental motion, timing, transitions, and what must stay stable.
Yes. A still image can become a keyframe. The prompt then adds movement, duration, continuity, and negative constraints for the video model.