Meta tags:

Headings (most frequently used words):

with, long, memory, relic, interactive, video, world, model, horizon, abstract, overview, walking, into, paintings, diverse, art, styles, term, multi, key, control, comparison, others, citation,

Text of the page (most frequently used words):
long (13), #memory (13), relic (9), and (9), with (8), video (7), world (7), interactive (6), horizon (6), drag (6), explore (6), model (6), yicong (5), hong (5), yiqun (5), mei (5), chongjian (5), models (4), our (4), into (4), the (4), term (4), more (4), real (4), time (4), research (3), work (3), following (3), control (3), distillation (3), teacher (3), student (3), using (3), self (3), for (3), three (3), these (3), adobe (2), yiran (2), yang (2), zhou (2), sai (2), yannick (2), hold (2), geoffroy (2), mike (2), roberts (2), matthew (2), fisher (2), eli (2), shechtman (2), kalyan (2), sunkavalli (2), feng (2), liu (2), zhengqi (2), hao (2), tan (2), please (2), recent (2), all (2), first (2), action (2), key (2), overview (2), compressed (2), second (2), bidirectional (2), autoregressive (2), forcing (2), efficient (2), training (2), tech (2), report (2), streaming (2), consistent (2), spatial (2), highly (2), this (2), that (2), given (2), enables (2), aware (2), duration (2), exploration (2), arbitrary (2), camera (2), retrieval (2), generation (2), online, relicworldmodel2025, author, title, url, https, worldmodel, github, year, 2025, you, find, inspiring, incorporate, your, consider, citing, citation, hunyuan, gamecraft, matrix, game, ours, compare, applying, identical, sequence, shown, column, evaluate, their, capability, scene, generalization, comparison, others, multi, diverse, art, styles, walking, paintings, pipeline, distilled, fast, enabled, through, replayed, back, propagation, see, details, truly, requires, ingredients, precise, user, however, most, existing, approaches, address, only, one, aspects, isolation, achieving, simultaneously, challenging, example, mechanisms, often, degrade, performance, present, unified, framework, tackles, challenges, altogether, single, image, text, description, scenes, built, upon, diffusion, techniques, represents, historical, latent, tokens, encoded, both, relative, actions, absolute, poses, within, cache, compact, structure, supports, implicit, content, enforces, coherence, minimal, computational, overhead, parallel, fine, tune, generate, sequences, beyond, its, original, transform, causal, generator, new, paradigm, full, context, over, well, rollouts, implemented, 14b, parameter, trained, curated, unreal, engine, rendered, dataset, achieves, fps, while, demonstrating, accurate, stable, robust, compared, prior, capabilities, establish, strong, foundation, next, modeling, abstract, starting, images, authors, equal, technical, contribution, project, lead,

Text of the page (random words):
interactive video world models relic interactive video world model with long horizon memory chongjian ge yicong hong yiqun mei chongjian ge yicong hong yiqun mei chongjian ge yicong hong yiqun mei chongjian ge yicong hong yiqun mei yiran xu yang zhou sai bi yannick hold geoffroy mike roberts matthew fisher eli shechtman kalyan sunkavalli feng liu zhengqi li hao tan first authors equal technical contribution project lead adobe research tech report drag to explore drag to explore interactive world exploration given arbitrary starting images abstract a truly interactive world model requires three key ingredients real time long horizon streaming consistent spatial memory and precise user control however most existing approaches address only one of these aspects in isolation as achieving all three simultaneously is highly challenging for example long term memory mechanisms often degrade real time performance in this work we present relic a unified framework that tackles these three challenges altogether given a single image and a text description relic enables memory aware long duration exploration of arbitrary scenes in real time built upon recent autoregressive video diffusion distillation techniques our model represents long horizon memory using highly compressed historical latent tokens encoded with both relative actions and absolute camera poses within the kv cache this compact camera aware memory structure supports implicit 3d consistent content retrieval and enforces long term coherence with minimal computational overhead in parallel we fine tune a bidirectional teacher video model to generate sequences beyond its original 5 second training horizon and transform it into a causal student generator using a new memory efficient self forcing paradigm that enables full context distillation over long duration teacher as well as long student self rollouts implemented as a 14b parameter model and trained on a curated unreal engine rendered dataset relic achieves real time generation at 16 fps while demonstrating more accurate action following more stable long horizon streaming and more robust spatial memory retrieval compared with prior work these capabilities establish relic as a strong foundation for the next generation of interactive world modeling overview overview of our long video distillation pipeline with compressed memory a 20 second bidirectional teacher is distilled into a fast autoregressive student model using self forcing and memory efficient training is enabled through replayed back propagation please see our tech report for more details walking into paintings drag to explore diverse art styles drag to explore long term memory drag to explore multi key control drag to explore comparison with others we compare relic with recent video world models by applying an identical control sequence to all models shown in the first column to evaluate their action following capability scene generalization and long term memory relic ours matrix game 2 0 hunyuan gamecraft 1 0 citation if you find our work inspiring or incorporate it into your research please consider citing the following online relicworldmodel2025 author yicong hong yiqun mei chongjian ge yiran xu yang zhou sai bi yannick hold geoffroy mike roberts matthew fisher eli shechtman kalyan sunkavalli feng liu zhengqi li hao tan title relic interactive video world models with long horizon memory url https relic worldmodel github io year 2025 adobe research