{"id":"da8bb37c-d22e-4a48-b470-1eb76752883f","arxiv_id":"2607.03732","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A training-free method preserves proxy-video dynamics via region-wise latent noising and Stochastic Flow Relaxation, outperforming editing and motion-transfer baselines on a custom 76-video set.","lead":"ProxyUp turns a coarse proxy video into a motion guide so a pretrained video model can keep that motion while inventing new scenes from text. It matters for anyone who needs controllable, physically plausible video without collecting paired training data.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.5","headline":"Prop. 1 only guarantees asymptotic erasure of proxy structure; finite-K SFR therefore depends on an unanalyzed incomplete-mixing regime to keep dynamics while cleaning OOD artifacts.","rationale":"The reader already isolated the finite-K SFR / Prop. 1 gap as the weakest assumption supporting the central claim. The present critique sharpens the same point: the proof is purely asymptotic and therefore predicts erasure of the very signal the method is supposed to preserve; success at K=15 is therefore an unanalyzed empirical compromise. The proposed optical-flow-vs-K sweep would quantify that compromise directly and either confirm or refute that the operating point still carries usable dynamics. Because the concern is already reflected in the reader’s CONDITIONAL verdict (and the paper’s own limitations section), no change of verdict is warranted; the contribution remains accept-shaped once artifacts and larger benchmarks appear.","tokens_in":15615,"tokens_out":624,"duration_ms":24843,"concrete_test":"On a fixed 10-video subset, measure masked optical-flow EPE (or RAFT correlation) between proxy foreground and generated foreground, plus IQ and MR, for K \\in {0,5,10,15,20,30} with all other hyperparameters fixed. If motion fidelity falls >20 % from K=0 to the paper’s K=15 while quality metrics rise, the simultaneous-preservation claim is an unprincipled trade-off rather than a robust outcome of Prop. 1.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The strongest claim (dynamics preservation + coherent prompt-aligned interactions via region-wise noising + SFR) rests on the finite-K=15 SFR loop (Sec. 4.4, Alg. 1 lines 6-10) taking the hand-composed latent Z^(0) of Eq. 5 close enough to the model manifold for visual quality and fg-bg coupling without erasing the inverted foreground that carries the proxy motion. Proposition 1 and App. A prove only that the re-noising kernel is ergodic and D_KL(q_K || \\pi_ID) \\to 0 as K \\to \\infty (under a well-trained v_\\theta), which would fully replace the proxy-conditioned latent by a pure text-conditioned sample at t_init—i.e., intermediate-noise T2V that forgets the proxy. No mask is re-applied inside the SFR iterations, no finite-K mixing-time or information-retention bound is given, and the operating point K=15 / t_init=0.922 is chosen purely by qualitative inspection (Fig. 6). Thus dynamics retention is an empirical side-effect of incomplete relaxation, not a consequence of the stated theory; if the residual proxy signal is already weak at that K, the outperformance on MR/Mech. (Table 1) cannot be attributed to the claimed mechanism.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper introduces proxy-conditioned video generation: a coarse proxy video (simulation or real recording) supplies foreground dynamics while a text prompt specifies novel content and interactions. Because paired proxy–target data are scarce, the authors propose ProxyUp, a training-free pipeline on pretrained rectified-flow video models (primarily Wan2.2). ProxyUp (i) ODE-inverts the masked proxy foreground to an intermediate noise level, (ii) composes a hybrid latent by region-wise latent noising (Eq. 5), (iii) applies Stochastic Flow Relaxation (SFR; Sec. 4.4, Alg. 1) to pull the hand-composed latent toward the model manifold, and (iv) finishes with deterministic ODE sampling. On a 76-clip custom set (simulation + real), ProxyUp reports gains over editing, inpainting, and motion-transfer baselines on Imaging Quality, Motion Rationality, Mechanics, and Material (Table 1), with qualitative and ablation support (Figs. 4–5).","tokens_in":15992,"tokens_out":1715,"duration_ms":18780,"significance":"If the empirical claims hold, the work offers a practical, training-free route to controllable dynamics that text alone cannot specify, and a useful intermediate between pure T2V, video editing, and motion transfer. Strengths include a clear three-stage pipeline, an explicit algorithm, an ablation isolating RLN and SFR (Fig. 5), hyperparameter sweeps (Fig. 6), cross-backbone checks in the appendix, and a formal (if asymptotic) Markov argument for SFR (Prop. 1, App. A). The task framing and the idea of using low-fidelity proxies as dynamics carriers are timely for physics-aware video generation. Significance is tempered by a small custom evaluation set, author-written VBench-style QA criteria, and a theory–practice gap on finite-K dynamics retention.","major_comments":[{"comment":"Sec. 4.4 / Prop. 1 / App. A: Proposition 1 only shows that the SFR Markov kernel is ergodic and D_KL(q_K || π_ID) → 0 as K → ∞ under a well-trained velocity field. That limit is pure text-conditioned sampling at t_init and would erase the inverted proxy structure. No mask is re-applied inside the SFR loop (Alg. 1 lines 6–10), and no finite-K mixing-time or information-retention bound is given. Dynamics preservation at the operating point K=15, t_init=0.922 (s=0.8) is therefore an empirical incomplete-mixing effect, not a consequence of the stated theory. The central mechanism claim—that RLN+SFR jointly preserve proxy dynamics while restoring fg–bg coupling—needs either a finite-K analysis (e.g., how much inverted foreground signal remains after K steps) or a clear reframing that SFR is a practical regularizer whose dynamics retention is empirical.","section":"Sec. 4.4, Prop. 1, App. A, Alg. 1"},{"comment":"Sec. 5.1 / Table 1 / App. C.2–C.3: The evaluation set has only 76 custom clips with author-written multi-question criteria for MR, Mech., and Mat. conditioned on the same proxy/prompt pairs used for generation. This is acceptable for a new task but is load-bearing for the claim of consistent outperformance in dynamic fidelity. Please (i) release the full metric prompts and scoring protocol as promised, (ii) report inter-annotator or multi-run variance, and (iii) add at least one external or human preference study on dynamics fidelity vs. text alignment so that Table 1 is not solely self-defined QA.","section":"Sec. 5.1, Table 1, App. C.2–C.3"},{"comment":"Sec. 5.2 / App. C.4: Several baselines run on different backbones and default schedules (DiTFlow on CogVideoX-5B; FlowDirector on Wan2.1; VACE 14B). Appendix cross-backbone checks (Figs. 9–10) and the SDEdit step-budget study (Fig. 8) help, but the main Table 1 still mixes generators. For the primary comparison, either re-run the strongest motion-transfer and editing baselines on the same Wan2.2 backbone used by ProxyUp, or report a backbone-matched subset as the headline table so gains on MR/Mech. can be attributed to the method rather than model capacity.","section":"Sec. 5.2, Table 1, App. C.4"},{"comment":"Sec. 5.4 / Fig. 6: Free parameters K, t_init (strength s), and CFG scales are chosen by qualitative inspection. Given that the skeptic concern is precisely the incomplete-mixing regime, please quantify the trade-off: e.g., proxy-motion metrics (optical-flow or keypoint correlation in the masked region) vs. K and s, not only visual examples. Without this, it is hard to know how fragile the reported MR/Mech. gains are to hyperparameter choice.","section":"Sec. 5.4, Fig. 6"}],"minor_comments":[{"comment":"Fig. 2 caption and body: the preliminary analysis is helpful; please state the exact strength/t values and masks used for each baseline so the trade-off narrative is reproducible.","section":"Sec. 3, Fig. 2"},{"comment":"Eq. (4): the background noise variance ((1−t_init)^2 + t_init^2)I is nonstandard relative to the linear path Z_t=(1−t)Z_0+tZ_1; a one-sentence justification (marginal variance of the interpolation) would help readers.","section":"Sec. 4.3, Eq. (4)"},{"comment":"Related work (Sec. 2.1) mentions physics-related conditions but cites little recent simulation-to-video or physics-prior work; a few additional pointers would better situate proxy videos among existing control signals.","section":"Sec. 2.1"},{"comment":"Limitation section (Sec. 6) and Fig. 11 are candid; consider moving one failure case into the main paper so readers see the dependence on the base model’s physical prior without opening the appendix.","section":"Sec. 6, Fig. 11"},{"comment":"Notation: strength s is defined in a footnote and reused as s=0.8; define it once in the main text near t_init for clarity.","section":"Sec. 4–5"},{"comment":"Minor typos / consistency: “V ACE” spacing in Table 1 and captions; “out-of-distribution (OOD)latent” missing space (Sec. 4.2); arXiv id and “Preprint” header are fine for review but should be cleaned for camera-ready.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The empirical story is plausible and the ablations are honest; the main risk for a top venue is overclaiming mechanism via Prop. 1 while the real operating regime is incomplete mixing, plus a small author-curated eval. I would accept after major revision if they (1) reframe or bound finite-K SFR, (2) strengthen evaluation transparency/human study, and (3) backbone-match the headline table. Not a reject: the task and training-free recipe are useful even if the theory is asymptotic only."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The useful bit is the task framing plus a concrete recipe: invert the proxy, keep the masked foreground latent, noise the rest, then run a short stochastic denoise/re-noise loop (SFR) before ODE sampling. That combination is cleaner than treating the problem as pure editing or pure motion transfer, and the preliminary analysis in §3 is honest about why those baselines fail on interaction synthesis.\n\nWhat works: region-wise latent noising is simple and does the job of decoupling appearance from motion. Ablations (Fig. 5) show both pieces matter. Table 1 and the qualitative examples give a consistent story that ProxyUp keeps proxy motion better than SDEdit/VACE/DiTFlow/FlowDirector while still regenerating prompt content. Cross-backbone checks in the appendix help a bit with the backbone mismatch. The method is training-free and the algorithm is fully specified.\n\nSoft spots, in proportion. The evaluation set is only 76 custom clips with author-written VBench-style QA criteria; that is fine for a methods paper but not a shared benchmark. Hyperparameters (K=15, s=0.8) are chosen by eye. The stress-test note is right on the theory: Prop. 1 only shows KL\to0 as K\to∞, which would erase the proxy. No mask is re-applied inside SFR, and there is no finite-K retention bound. Dynamics preservation is therefore an empirical incomplete-mixing effect, not a consequence of the stated Markov argument. That does not kill the method—it still works in the figures—but it means the paper over-claims the theory relative to the operating point. Also, success still depends on the base model already knowing the interaction (Newton’s cradle failure is acknowledged).\n\nWho it is for: people building sim-to-video or controllable animation pipelines who need a practical training-free lever. Worth a serious referee. I would engage, cite the task framing and the recipe, and treat the asymptotic proof as optional decoration.","headline":"Clean training-free recipe for proxy-as-dynamics; finite-K SFR is the real mechanism and the asymptotic proof does not explain it.","tokens_in":16621,"tokens_out":487,"would_cite":true,"duration_ms":4938,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"Coarse proxy videos can steer complex motion in pretrained video generators without any training, by noising latents region-wise and relaxing them onto the model manifold.","keywords":["proxy-conditioned video generation","training-free video synthesis","region-wise latent noising","stochastic flow relaxation","controllable dynamics","rectified flow","motion transfer","video editing"],"falsifier":"If, on the same backbone and equal or larger inference budget, a region-wise-noised SDEdit or editing baseline matches or exceeds ProxyUp on Motion Rationality and Mechanics for the bread-cutting and curtain-pulling proxies (and similar held-out dynamics), the claimed benefit of SFR would not hold.","tokens_in":16536,"feed_emoji":"🎬","tokens_out":859,"duration_ms":12044,"temperature":0.7,"pith_summary":"Modern text-to-video models still struggle to specify fine-grained, physically plausible motion and interactions from language alone. This paper introduces proxy-conditioned video generation: a coarse video from simulation or real recording acts only as a dynamics carrier for foreground motion, while a text prompt drives new content and scene interactions. Because paired proxy–target videos are hard to collect, the authors propose ProxyUp, a training-free pipeline on pretrained flow-based video models. It inverts the proxy, keeps motion-critical latents in a masked region, injects noise elsewhere for regeneration, then runs Stochastic Flow Relaxation to pull the hybrid latent toward the model’s learned distribution before ODE sampling. On both physics-simulation and real-world proxies, the method improves dynamic fidelity and text alignment over strong editing and motion-transfer baselines.","feed_headline":"Proxy videos steer video motion without retraining","feed_subtitle":"Region-wise noising and flow relaxation keep dynamics while text rebuilds the scene.","key_machinery":"Region-wise latent noising plus Stochastic Flow Relaxation (SFR): invert the masked proxy to an intermediate noise level, keep those latents in motion-critical regions while replacing the rest with matched noise, then iteratively denoise and re-noise the hybrid latent so it approaches the model’s in-distribution manifold before deterministic ODE sampling.","core_discovery":"The paper claims that a training-free combination of region-wise latent noising and Stochastic Flow Relaxation lets a pretrained video generator preserve essential dynamics from a coarse proxy video while synthesizing novel, prompt-aligned content and plausible foreground–background interactions—outperforming video editing and motion-transfer baselines on dynamic fidelity and text alignment for both simulated and real proxies.","pith_inferences":["The same recipe could turn low-fidelity game or robotics rollouts into large synthetic video corpora with controlled physics and varied visual styles.","When the base generator lacks the needed interaction priors, better proxies alone will not fix failures—pointing toward models trained on richer physical contact data.","Automatic or learned masks and force-application cues would reduce reliance on SAM-style foreground masks and hand-chosen t_init and K."],"forward_implications":["Physics simulations and casual real recordings can be reused as motion controllers for open-ended text-driven video synthesis without collecting paired training data.","Video editing and motion-transfer pipelines that stay anchored to source appearance are not the right tools when the source is only a dynamics prior.","A modest inference-time relaxation loop can repair hybrid latents that would otherwise break foreground–background coupling.","Task-specific proxy–prompt–mask evaluation sets become necessary to measure dynamics-preserving regeneration rather than pure editing or pure generation."],"fun_headline_variants":["Proxy videos control dynamics in pretrained models without training","Region-wise noising and flow relaxation lock proxy motion to text","Training-free proxies preserve motion while regenerating scenes","Invert proxies then noise selectively to steer video generation","SFR relaxes composed latents so proxies drive prompt-aligned videos"],"cache_read_input_tokens":128,"weakest_assumption_plain":"That a finite number of SFR re-noising rounds is enough to move a hand-composed, out-of-distribution latent onto the pretrained model’s manifold so that sampling invents coherent interactions the base model already knows how to draw.","fun_headline_variants_meta":{"raw":{"variants":["Proxy videos control dynamics in pretrained models without training","Region-wise noising and flow relaxation lock proxy motion to text","Training-free proxies preserve motion while regenerating scenes","Invert proxies then noise selectively to steer video generation","SFR relaxes composed latents so proxies drive prompt-aligned videos"]},"model":"grok-4.5","effort":"low","cost_usd":0.004626,"raw_usage":{"total_tokens":1347,"prompt_tokens":768,"num_sources_used":0,"completion_tokens":64,"cost_in_usd_ticks":46260000,"prompt_tokens_details":{"text_tokens":768,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":515,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":768,"tokens_out":64,"duration_ms":4372,"temperature":1.0,"reasoning_tokens":515,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-12T00:18:51.928688+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"If, on the same backbone and equal or larger inference budget, a region-wise-noised SDEdit or editing baseline matches or exceeds ProxyUp on Motion Rationality and Mechanics for the bread-cutting and curtain-pulling proxies (and similar held-out dynamics), the claimed benefit of SFR would not hold.","supporting_citations":[],"review_version":1}