Meta tags:

Headings (most frequently used words):

object, motion, editing, with, comparison, first, frame, path, baseline, trace, in, videos, trajectory, guidance, teaser, our, results, design, multiple, designs, for, same, input, full, pipeline, baselines, applications, zero, shot, multi, cross, view, transformation, video, resynthesis, bibtex, insertion, control, replacement,

Text of the page (most frequently used words):
video (23), motion (13), the (13), input (12), frame (11), #object (10), first (10), path (8), edited (7), trace (6), and (6), editing (5), with (4), #trajectory (4), videos (3), arxiv (3), ours (3), prompt (3), comparison (3), adding (3), step (3), guidance (2), phung (2), quynh (2), mai (2), long (2), ham (2), cusuh (2), liu (2), feng (2), huang (2), jia (2), bin (2), mahapatra (2), aniruddha (2), bibtex (2), gencompositor (2), vace (2), hunyuan (2), pisco (2), canvas (2), wan (2), move (2), magic (2), resynthesis (2), baseline (2), based (2), warping (2), cross (2), view (2), transformation (2), child (2), rainbow (2), for (2), our (2), article, phung2026trace, title, author, journal, preprint, 2603, 25707, year, 2026, copy, masked, depthanything, megasam, interpolation, zero, shot, multi, kite, airplane, buffalo, tiger, penguin, replacement, wearing, headset, turtle, smoke, man, deer, insertion, control, applications, full, pipeline, baselines, multiple, designs, same, loading, prompts, results, design, model, generates, final, moving, along, specified, while, preserving, visual, quality, original, seamlessly, inpainting, region, where, previously, appeared, module, converts, into, dense, per, correspondences, across, all, frames, accounting, camera, given, user, draws, specify, desired, teaser, pdf, university, maryland, college, park, adobe, research,

Text of the page (random words):
trace object motion editing in videos trace object motion editing in videos with first frame trajectory guidance quynh phung 1 long mai 2 cusuh ham 2 feng liu 2 jia bin huang 1 aniruddha mahapatra 2 1 university of maryland college park 2 adobe research pdf arxiv bibtex teaser step 1 given an input video the user draws a trajectory on the first frame to specify the desired object motion step 2 our cross view motion transformation module converts the first frame trajectory into dense per frame motion correspondences across all video frames by accounting for the camera motion in the video step 3 a video resynthesis model generates the final edited video moving the object along the specified path while preserving the visual quality of the original input video and seamlessly inpainting the region where the object previously appeared our results first frame path design loading prompts input video first frame path edited video multiple path designs for same input input video prompt first frame path edited video full pipeline comparison with baselines input video first frame path prompt magic motion wan move motion canvas pisco hunyuan vace gencompositor trace ours applications object insertion with motion control adding a deer adding a child adding a man input video edited video object editing rainbow smoke rainbow turtle wearing vr headset input video edited video object replacement child penguin buffalo tiger kite airplane input video edited video zero shot multi object editing input video first frame path edited video cross view motion transformation baseline comparison input video first frame path interpolation megasam based warping depthanything v3 based warping trace ours video resynthesis baseline comparison input video masked video prompt magic motion wan move motion canvas pisco hunyuan vace gencompositor trace ours bibtex copy article phung2026trace title trace object motion editing in videos with first frame trajectory guidance author phung quynh and mai long and ham cusuh and liu feng and huang jia bin and mahapatra aniruddha journal arxiv preprint arxiv 2603 25707 year 2026