Motion Control

Motion Control copies motion from a reference video onto a still image. You keep the face, costume, or character from the photo, and you keep the performance from a real clip — a dance, a wave, a walk, or a product turn. The generator rebuilds a video so the subject follows that motion instead of inventing movement from a text prompt. Unlike a generic image-to-video field, the motion here is specified by footage you already have.

Creators reach for this when they already like how someone looks in a photo and already like how someone else moves on camera. Reshooting both at once is expensive. Two uploads compress that loop: identity and styling from the image; timing, path, and energy from the reference video. Preview and download stay on the same URL.

This page is the workspace, not a teaser. The form above accepts a JPEG or PNG still (max 10MB, longer side above 300px, aspect ratio between 2:5 and 5:2) and an MP4, MOV, or MKV reference video (max 100MB, 3–30 seconds). Optional prompt text can nudge scene details. Character orientation chooses whether the output follows the pose of the image (up to about 10 seconds) or the framing of the video (up to 30 seconds). Resolution is 720p or 1080p. Credits show before you generate.

Treat it as a practical motion-transfer pass. It will not replace a motion-capture stage, and it will not invent limbs that were cropped out of the still. A sharp character, a readable performance, and matching framing give the model something to attach motion to. If the still is a tiny figure in a dark room, or the reference video is a montage of jump cuts, there is little continuous motion to copy.

Signup credits cover first runs so you can learn the loop before buying packs. Billing scales with duration and resolution: 12 credits per second at 720p and 18 credits per second at 1080p. Privacy settings let you keep a clip private or apply copyright protection. Generate, wait, then download — all on this Motion Control URL.

Character orientation is the main lever after the two files. Match Image is for portraits and product heroes that should keep the still’s facing. Match Video is for dance, walk cycles, and full-body motion where the reference clip already shows how the body should turn. Pick the mode that matches the performance you uploaded, not a default you never revisit.

Duration follows the reference video, within the mode’s cap. A 20-second dance in Match Video can become a longer take; the same clip in Match Image will not magically keep 20 seconds if that path is capped near 10 seconds. Trim the reference to the beat you actually want copied. Jump cuts and shaky handheld footage make the model guess between frames.

Resolution is a publish choice, not a creativity slider. 720p is enough to judge whether the character locked. 1080p is for the take you will post. Optional prompts can mention wardrobe or background, but they should not fight the reference video. If the clip is a wave and the prompt says “sprinting,” the motion still comes from the wave. Keep prompts short.

Give the model a subject it can hold. A well-lit character filling the frame beats a distant figure. If the still is noisy, cropped through the joints, or facing away, recover a better photo first. A second arm that was never in the image cannot be invented.

Match framing between still and reference video. Pair a head-and-shoulders photo with upper-body motion. Pair a full-body still with dance or walk cycles. Mixed framing is the usual reason a take looks melted. One subject in both files is safer than a crowd.

Keep the reference video boring in a good way: continuous action, moderate speed, stable camera. Then change one setting per retry — orientation or resolution — so you know which lever worked. Download when the motion reads, then use related Pixwit tools if you still need an avatar, a different video model, or a longer generated shot.