betaFeedback

WorksComputer Science

Can World Models Predict Rice Growth? Toward Agricultural World Models

Abstract

World models can imagine how a physical scene evolves over the next few seconds, but growers need to know how a crop will develop over weeks to months: will this rice plant have headed a month from now, and how will it look at harvest? We introduce agricultural world modeling, the task of predicting how the same plant will look at a requested horizon of days to months, from one photo of it, its cultivar, and its camera view. In preliminary analysis, even state-of-the-art image generators get this wrong: they often grow the plant too little or give it the color and form of the wrong growth stage. We argue that they do not work out the plant's developmental state, which the photo reveals only indirectly. We therefore propose AgroWM, which reasons about development before it draws growth. Its reasoning module, PhenoClock, works out the current growth stage, predicts the future one, and writes a Growth Script with numeric targets for height, plant area, and color; its generation module, the Rooted Renderer, then edits the photo to follow this script, adding new organs and colors to the photographed plant. Because pixel scores can prefer an unchanged copy of the photo, we instead check each forecast's traits and developmental timing against the plant's own later photos. On rice, 54.6% of the forecasts of AgroWM are matched to the target stage (Stage R@1), against 34.8% for GPT Image 2.5, 24.0% for Qwen Image 2.1, and 47.4% for the strongest domain-specific baseline, the same backbone fine-tuned without Growth Scripts; 9 to 20 weeks ahead, it still reaches 46.4%, against 35.7% for that baseline. On maize, a crop with a different body plan and stage scale, the same design again scores best on all four measures.

Introduction

World models [1, 2, 3] let machines imagine how a physical scene evolves over the next few seconds, as when a glass falls, strikes a surface, and shatters (Figure 1a). Questions about growing food instead require forecasts over weeks to months: how tall will this plant grow, how much will its canopy spread, and when will it flower and fill its grain? For rice, which feeds more than half of the world’s population [4], the answers otherwise take up to a year to arrive. We ask whether a world model can give them in advance: from one photo of a plant, its cultivar, and its camera view, predict how the same plant will look at a requested horizon of days to months (Figure 1b). We call this task agricultural world modeling. Its answer is a photo, which, unlike a single number, can be checked by eye and measured for any visible trait.

Conventional and agricultural world models

Figure 1. From physical dynamics to growth. (a) A conventional world model predicts rapid physical events. (b) From one photo, an agricultural world model predicts how the plant will look at each requested horizon of weeks to months, as it moves through growth stages.

The task is harder than it looks, as shown by a preliminary study of 64 rice cases: 1–4 weeks (22 cases), 5–8 weeks (24), and 9–20 weeks (18) (§ 3). First, two state-of-the-art image generators fail at it: when Image 2.1 hallucinates a 1-1 radically changes the plant, although real plants grow about tenfold in projected area over 9 to 20 rounded weeks (median 11.2×), and GPT Image 2 grows it too little at long horizons, shows senescence-related color changes at the wrong stage, and redraws it toward a generic rice plant (Figure 2). Second, the horizon alone does not decide the future stage: of the 18 cases with horizons of 9 to 20 weeks, 6 plants have not yet headed by the target date, while 9 are already filling grain or ripening. A forecaster must therefore infer each plant’s developmental state, which the photo reveals only indirectly. Third, common image scores reward doing nothing: a copy of today’s photo has the highest SSIM in Table 1, and Qwen Image 2.1, which barely changes the plant, the highest PSNR.

Figure 2. How existing models fail to forecast growth, and why standard scores miss it.

MethodDINO↑IoU↑Area↓Height↓Density↓Yellow↓Stage R@1↑Event R@1↑PSNR↑SSIM↑
General-purpose
GPT Image 20.7090.3035.5911.496.3124.4035.312.813.230.689
GPT Image 2.50.7210.2996.9212.576.7029.1734.810.812.920.694
Qwen Image 2.10.6930.2417.8518.286.8332.3124.01.114.200.748
Qwen-Image-Edit-25110.5680.18828.8420.4625.0833.7626.59.29.540.476
FLUX.2 Klein0.4960.25941.5822.4736.5134.1125.29.67.360.392
FLUX 3 Image Klein Base0.6830.24413.8717.6512.3835.1424.11.412.980.691
LANCE0.6360.24023.8120.9024.1731.6124.53.810.550.591
Domain-specific
Copy0.7020.2527.1719.066.2429.3323.20.014.020.750
LANCE FT0.8150.3423.918.045.6215.9546.916.913.940.709
FLUX.2 Klein FT0.7670.3054.7011.006.0324.5939.318.013.910.697
Qwen Image 2.1 FT0.8340.3233.737.835.3311.4347.422.514.100.685
Agricrafter0.6660.3295.209.617.9715.7932.113.314.110.706
Ours
AgroWM0.8600.3273.517.434.9810.0054.624.814.140.686

Table 1. Main comparison on rice. Bold: best per column; PSNR and SSIM for reference only.

A reliable growth forecast must get the individual plant, its future traits, and their timing right. Existing methods address only parts of this problem. Time-lapse generators [5, 6] draw plausible growth from text, with no particular plant or date to match, and keyframe interpolation [7] and 4D reconstruction [8] are given the later images that a forecast has to predict. Plant-growth GANs [9, 10, 11] come closest, but they map elapsed time, treatment, or simulated biomass to pixels without reasoning about development. Agricrafter [12] generates controllable videos of a crop’s full growth cycle, but a coherent sequence alone does not show whether a given plant reaches the correct stage within the requested horizon. Crop models [13, 14] do track development using temperature and day length, but require cultivar- and site-specific parameters and do not directly generate the future appearance of an individual plant. Our approach follows an agronomist’s reasoning by first predicting the plant’s future developmental stage and then generating an image consistent with that prediction. For example, if a plant is currently tillering and is expected to reach jointing in four weeks, the forecast image should show elongated stems and a taller plant.

We turn this insight into AgroWM, an agricultural world model with two modules (Figure 3). First, its reasoning module, PhenoClock, is a vision-language model that works out the plant’s current stage, predicts its stage at the horizon, and writes this chain of thought (CoT) as a Growth Script. The script is a short, structured note: the current and future stage and, for each visible trait such as height, plant area, and color, today’s value, the target value, and the change in words. Since no dataset records such notes, PhenoClock learns from scripts written in hindsight: for each training pair, the stages and values come from records and measurements of the real future, and a language model adds the words (§ 5.2). Second, its generation module, the Rooted Renderer, then edits today’s photo as the script says; it is prompted and trained to keep the plant’s form, framing, and scale, on pairs chosen to reward real growth (§ 5.3).

Figure 3. AgroWM forecasts development before rendering growth. PhenoClock writes a Growth Script with the future stage and targets for height, plant area, and color, and the Rooted Renderer edits the source photo to follow it. Example: a validation pair 49 days apart (7 weeks in the prompt), with the Hindsight Tutor’s reference script (center) and the forecast from PhenoClock’s own script (right).

To judge forecasts, we check what matters to a grower: the same plant, at the right size and color, at the right stage of development (§ 4.1). Timing is checked against the plant’s own photo series: Stage R@1 is the share of forecasts whose most similar photo of the plant lies in the target stage. On held-out plants of the public rice dataset WGSR [15], AgroWM reaches 54.6% Stage R@1, against 34.8% for GPT Image 2.5 and 24.0% for Qwen Image 2.1, two state-of-the-art generators, and 47.4% for the strongest domain-specific model, the same backbone fine-tuned on the same pairs without Growth Scripts (Table 1). Reasoning about development thus adds 7.2 points (95% CI 5.2–9.2), and 10.7 points 9 to 20 weeks ahead (Figure 4). On maize, which grows a single stalk with a tassel instead of tillers and panicles, the same design again scores best, so the approach is not tied to one crop (§ 6.3).

Figure 4. Forecast quality by horizon on rice.

In summary, we make three contributions:

  • We explore agricultural world modeling and find that the current image generators do not pace a plant’s growth to the horizon, reach its target stage, or keep the individual.

  • We propose AgroWM, which reasons before it renders: PhenoClock forecasts the developmental stage with the plant’s height, area, and color as an agricultural CoT learned from scripts written in hindsight, and the Rooted Renderer turns this forecast into an edited photo of the plant; it gives the best forecasts on both rice and maize.

  • We show that standard metrics such as PSNR and SSIM are useless in guiding the generation of ecologically meaningful plants, and introduce an evaluation of identity, geometry, color, and timing against the plant’s growth series.

Related Work

World Models in Computer Vision

World models and long-horizon prediction. World models learn how an environment evolves, first as latent simulators for control [16, 17] and now as large video generators presented as world simulators [1, 18, 19] for games [20, 21], driving [2], and robots [3, 22]. Longer horizons have been reached by temporal abstraction, which predicts directly at coarse or adaptive steps [23, 24, 25, 26, 27], and learned weather models forecast physical fields days ahead [28, 29]. These models primarily predict physical dynamics, including object motion, interactions, and the evolution of physical fields. We open a regime they do not cover: the growth of an organism over days to months, driven by a state that the image shows only indirectly.

Probing and evaluating world knowledge. Video generators often violate physical laws despite realistic frames [30, 31, 32] and lack a consistent time scale [33]. Time-lapse generation [5, 6, 34, 35] targets slow processes from text, with no individual’s real future as ground truth, keyframe interpolation [7] is given the last frame, and face aging [36] involves no new organs. These tasks often use pixel and perceptual metrics [37, 38, 39] to measure visual similarity, but such metrics can favor blurry averages under uncertainty [40, 41]. In addition, FID [42] compares image distributions and reflects both visual fidelity and diversity, but does not separate their contributions to the score. We extend this probing from physical consistency to the timing of biological development. Genotype, environment, and management shape a plant’s developmental trajectory, so elapsed time alone does not determine its future stage. A forecaster must infer where the plant is in its development and how far it will progress over the requested horizon. A realistic image can still be the wrong forecast: its size, color, or organs may be plausible but mistimed. We therefore score timing by retrieval in the same plant’s photo series (§ 4.1).

Reasoning before generation. Multimodal language models have been used to plan image edits [43, 44] and to reason before synthesis [45], extending chain-of-thought prompting [46], and CausalWM [47] reasons about causal variables before predicting video. Their reasoning concerns layout, semantics, or short-term physics. PhenoClock differs in what it reasons about and where its supervision comes from: it advances a hidden developmental state by a requested horizon and states it in image measurements, in a schema designed with agronomic knowledge. Its training scripts are written backwards from real future images around recorded stages and measured values, in the spirit of privileged information [48, 49] and rationalization [50].

Modeling Growth in Agriculture

Process-based crop and plant models. Agronomists have long forecast crops with process-based models: crop models such as ORYZA2000 [13], DSSAT [14], APSIM [51], and WOFOST [52] advance a developmental state with temperature and day length and simulate biomass and yield from weather, soil, and structural plant models simulate architecture [53, 54]. They need cultivar- and site-specific parameters and output numbers or geometry rather than the appearance of a particular plant. We keep their central idea, a developmental state advanced through time, but infer it from the photo and render it as an image.

Observing and forecasting growth from images. Image-based phenotyping describes the plant’s current state: vision models segment organs, measure traits, and estimate plant age from photographs [55, 56, 57]. Growth forecasting goes further by predicting its later appearance. Plant-growth GANs predict future images from earlier photos, using fixed prediction intervals or explicit time, treatment, and biomass conditions, and assess their outputs through image quality and trait agreement [9, 10, 11]. Agricrafter [12] generates full-growth-cycle crop videos with diffusion models, and 4D reconstruction [8] recovers growth that has already been imaged. We explicitly forecast the future developmental state of the plant to guide image generation from a current photo at horizons of up to 20 weeks, and evaluate the developmental timing of the plant’s photo series, on two crops.

Preliminary Analysis

We first check how state-of-the-art image generators forecast growth and whether standard scores reflect their mistakes. We test two of them: GPT Image 2, the general-purpose model with the highest Stage R@1 in Table 1, and Qwen Image 2.1, the base of our Rooted Renderer. The 64 validation cases come from 64 cultivars, half in each view, with 22, 24, and 18 cases at 1–4, 5–8, and 9–20 rounded weeks (the longest is 15 weeks; Figure 2). Each receives today’s photo and a prompt with the cultivar, view, and horizon, plus a keep-list that asks to keep the plant’s identity and to avoid premature yellowing or panicles.

How Existing Models Fail

① Growth does not keep up with the horizon. The further ahead a forecast looks, the more the plant should change, but neither generator keeps pace with the real change. Qwen Image 2.1 barely changes the plant: its median projected area is 0.93×0.93\times to 0.98×0.98\times today’s, while real plants grow 2.7×2.7\times, 4.4×4.4\times, and 11.2×11.2\times in the three ranges. GPT Image 2 grows the plant with the horizon but falls further behind, reaching 2.4×2.4\times, 3.0×3.0\times, and 5.6×5.6\times (Figure 2a).

② Forecasts often show the wrong growth stage. How a plant looks depends on its growth stage, and the generators often draw the wrong one. Although the prompt forbids it, GPT Image 2 yellows young plants early: in 8 of the 15 front-view cases that stay vegetative, the share of yellow or brown plant pixels in its forecast exceeds the real plant’s by more than 10 percentage points (Figure 2b). Both models keep ripening plants green: in the 22 cases at grain filling or ripening, the real plants are 54% yellow or brown on average, against 24% for GPT Image 2 and 23% for Qwen Image 2.1 (② in Figure 2). The horizon alone cannot fix the stage: at 9–20 rounded weeks, 6 of the 18 plants have not yet headed, while 9 are filling grain or ripening, so each plant follows its own clock.

③ Forecasts often show a different plant. A forecast must show the same individual plant, but GPT Image 2 often changes its form toward a typical rice plant. In front views it draws a taller, upright plant, more than 10 points of image height too tall in 13 of the 32 cases (Figure S1); in top views it draws a symmetric rosette (Figure 2a, Figure 2c). Qwen Image 2.1 keeps the individual mostly by not changing it.

Figure S1. Figure S1. How existing models fail to forecast growth, two single cases per failure, front and top views. Red frames mark failures, and the images of a case share one crop. Right: statistics over all 64 cases; ④ counts the cases in which each score ranks no growth or GPT Image 2 best.

Together, these failures suggest that the generators jump from today’s photo and the horizon straight to a new photo, without working out the plant’s developmental state, and fall back on a generic picture of rice. Our method is therefore built around three requirements: (T1) advance the developmental state explicitly (①); (T2) read the plant’s own stage first and derive its traits from it (②); and (T3) keep the individual and change only what development changes (③).

Our Evaluation Design

Assessing whether forecasts meet these requirements calls for more than visual similarity. We first examine where standard scores fall short, then define an evaluation of identity, size, color, and developmental timing.

④ Standard scores reward no growth. Standard scores do not catch these failures: pixel scores favor an unchanged photo, and the other scores each see only part of the change. In the 64 cases, plants cover only 14% of the real future photo, so PSNR and SSIM mostly measure the white background, and an unchanged copy of today’s photo gets the highest mean PSNR and SSIM of all forecasts we score (14.9 dB and 0.81, against 14.2 dB and 0.75 for GPT Image 2). DINOv2 similarity [39] depends less on the background, but its scale shifts with the horizon: the copy scores 0.74 at 1–4 rounded weeks and 0.41 at 9–20, so a value means little without its horizon. Single traits see growth but not development: GPT Image 2 beats the copy and Qwen Image 2.1 in projected area in 52 of the 64 cases, yet in 33 of them the forecast looks most like a photo from another stage (Figure 2b).

What must a forecast get right? We score four parts of a forecast separately. (i) Identity (③): DINOv2 feature similarity (DINO) and foreground-mask IoU with the real future. (ii) Size and shape (①, ③): errors, in points of the image, of projected area, plant height, and density, the plant’s share of its convex hull. (iii) Color (②): error of the yellow–brown share of plant pixels. (iv) Timing (①, ②): we rank the photos of the same plant and view, today’s included (14 on average), by DINOv2 ViT-B/14 similarity to the forecast; Stage R@1 and Event R@1 are the shares of forecasts whose top photo lies in the target stage or is the real target. Since these photos share the plant and camera, a forecast that does not grow matches today’s photo: Qwen Image 2.1 reaches an Event R@1 of only 1.1%, worse than picking a photo at random (at least 7%; Table 1). On the 2,000 validation queries, PSNR and SSIM prefer this no-growth forecast to AgroWM on 51% and 93%, the timing scores on 6% and 1% (Table S1; ④ in Figure 2), so PSNR and SSIM are for reference only. A Growth Script states the target size and color, so these errors may favor AgroWM; the timing scores use only the forecast image and recorded labels.

ScoreNo growth wins (%) CopyNo growth wins (%) QwenValidation mean QwenValidation mean AgroWM
PSNR ↑45.550.814.2014.14
SSIM ↑89.993.10.7480.686
DINO sim. ↑25.718.70.6930.860
Area err. ↓31.325.67.853.51
Height err. ↓29.629.118.287.43
Density err. ↓45.041.26.834.98
Yellow err. ↓28.821.332.3110.00
Stage R@1 ↑5.45.624.054.6
Event R@1 ↑0.00.61.124.8

Table S1. Which scores reward no growth? Left: share of the validation queries on which a score strictly prefers the copy or Qwen Image 2.1 to AgroWM; ties count in the denominator but not as wins. Right: validation means (errors in percentage points, R@1 in %). Height on 1,000 front views, DINO and R@1 on 1,964 queries, the rest on 2,000.

Method

AgroWM answers two questions in turn, what the plant will become and how it will then look (Figure 3), instead of asking one generator to guess the developmental state, the amount of growth, and what to keep all at once (§ 3). The reasoning module, PhenoClock (§ 5.2), answers the first: from today’s photo and the query, it predicts a measured Growth Script (T1, T2). The generation module, the Rooted Renderer (§ 5.3), answers the second: it edits the same photo as the script says (T3). In training, a Hindsight Tutor that sees both photos of a pair writes the words of its reference script around recorded stages and measured values.

Task Definition

The model sees one photo and predicts the plant at a requested future horizon. Given a photo xtx_t of a plant at time tt, a horizon Δ\Delta, and conditions c=(a,v)c=(a,v), the cultivar aa and camera view v∈{front,top}v \in\{\text{front},\text{top}\}, an agricultural world model predicts the same plant at time t+Δt+\Delta:

x^t+Δ=f(xt,Δ,c),1≤Δ≤T,(1)\hat{x}_{t+\Delta}=f(x_t,\Delta,c), \qquad1 \le\Delta\le T, \tag*{(1)}

where TT is the rest of the season. In the rice experiments, elapsed intervals are rounded to whole weeks for the model input, while target photos retain their recorded observation dates. The model must infer the current stage from the photo, since no stage label is given. The recorded stages and slot values (Table S2) of a pair (xt,xt+Δ)(x_t, x_{t+\Delta}) are available only in training and for evaluation. AgroWM splits ff into its two modules,

SlotViewDefinitionUnit
CulmFplant height (bottom edge to top plant pixel)% of image height
Leaves/areaF, Tall plant pixels / image pixels% of image
Plant densityFplant pixels / pixels of their convex hull%
CanopyTplant pixels / pixels of their convex hull%
PanicleF, Ton or after the accession’s heading date0/1
Color developmentF, Tyellow and brown / plant pixels, HSV thresholds%

Table S2. Slots of a Growth Script, image proxies measured on the plant pixels of the white-background photo (F: front, T: top view). Bottom: one growth-plan line of a Hindsight Tutor script, validation pair, tillering to jointing in 27 days.

s=g(xt,Δ,c),x^t+Δ=r(xt,Δ,c,s),(2)s = g(x_t, \Delta, c), \qquad\hat{x}_{t+\Delta} = r(x_t, \Delta, c, s), \tag*{(2)}

where PhenoClock gg writes the Growth Script ss and the Rooted Renderer rr edits the source photo accordingly.

Reasoning Module: PhenoClock

PhenoClock answers the first question, what the plant will become. It is a vision-language model whose output, a structured chain of thought, spells out how the plant will develop. Unlike free-form reasoning before generation [45], this chain of thought must be exact enough to draw and to check, and WGSR does not provide it. We therefore define what it contains, write training examples in hindsight, and teach PhenoClock to produce them from today’s photo.

Measured chain of thought. Like an agronomist’s note, a Growth Script states the stage first and the traits second, in three blocks of fixed order,

s=(zt,zt+Δ;{(dk,mtk,mt+Δk)}k;{ek}k),(3)s = (z_t, z_{t+\Delta}; \{(d_k, m_t^k, m_{t+\Delta}^k)\}_k; \{e_k\}_k), \tag*{(3)}

where kk runs over the slots Kv\mathcal{K}_v of view vv: culm height, plant area, density or canopy, panicles, and color (Table S2). The stage block gives today’s and the future stage on eight stages S1–S8 adapted from the IRRI scale [58] (tillering, jointing, booting, heading, grain filling, milk and dough, yellow ripening, and maturity); the growth-plan block gives, for each slot, a sentence dkd_k on the visible change with today’s and the target value; and the target-phenotype block describes the end state (eke_k). Read in order, a Growth Script locates the plant, advances it by Δ\Delta, and then observes, quantifies, and describes each slot. All values except the panicle label, which follows the cultivar’s heading date, are image measurements, such as the plant’s extent from the bottom edge and its share of the image: each is a target for PhenoClock, a condition for the Rooted Renderer, and a quantity measured again on the forecast (§ 4.1).

Scripts written in hindsight. We write the training scripts backwards, from pairs whose future we know. A language model explains well but measures poorly, so the Hindsight Tutor hh sees both photos and the facts of a pair, the recorded stages z=(zt,zt+Δ)z = (z_t, z_{t+\Delta}), the interval, and the slot values m=(mt,mt+Δ)m = (m_t, m_{t+\Delta}), but writes only the words ww, in a strict schema; a fixed template Π\Pi then inserts the stages and values and computes each change:

w=h(xt,xt+Δ,Δ,z,m),s∗=Π(w,z,m).(4)w = h(x_t, x_{t+\Delta}, \Delta, z, m), \qquad s^* = \Pi(w, z, m). \tag*{(4)}

The stages and numbers of s∗s^* are thus copied from the records and measurements. Because hh sees the answer, its prompt allows words only on what a forecaster could know: the kind and direction of change, organ emergence consistent with the panicle value, and color progression, never numbers, dates, organ counts, or future poses (leakage control). A validator checks the structure, the stages against the records, and forbidden forms such as numbers in the words, and hh rewrites rejected scripts (verification).

Forecasting from today’s photo. At test time the future is unknown, so PhenoClock must write the same Growth Script from today’s photo, the horizon, and the conditions alone. We fine-tune PhenoClock with a next-token loss on s∗s^* whose tokens fall into three groups, the stage names, the metric fields, and the remaining script text:

Lclock=0.1Lstage+0.4Lmetric+0.5Ltext,(5)\mathcal{L}_{\mathrm{clock}} = 0.1\mathcal{L}_{\mathrm{stage}} + 0.4\mathcal{L}_{\mathrm{metric}} + 0.5\mathcal{L}_{\mathrm{text}}, \tag*{(5)}

where each term is the loss −log⁡pg(si∗∣s<i∗,xt,Δ,c)-\log p_g(s_i^* \mid s_{<i}^*, x_t, \Delta, c) over the tokens si∗s_i^* of one group. The future reaches PhenoClock only through this label, as in learning with privileged information [48, 49], and the label’s source-side stage and values also teach PhenoClock to read today’s plant.

Generation Module: Rooted Renderer

The Rooted Renderer answers the second question, how the plant will then look: it should add the organs and colors that the Growth Script states and keep the rest of today’s photo, whereas the generators of § 3 either leave the plant unchanged (①) or alter it (③). We build it from a pretrained image generation and editing model with three choices, one per paragraph below: it edits today’s photo, it learns from pairs that reward real growth, and it trains on the Hindsight Tutor’s scripts.

Keeping the same plant. The Rooted Renderer edits today’s photo instead of drawing a new one: the source photo is the edit image in training and at test time, and every target keeps the frame of its source, in which the numbers of the Growth Script are defined. The prompt states the task (accession, view, and horizon in whole weeks), a fixed keep-list (same identity, background, framing, and scale; development only forward, without premature yellowing or panicles), and the Growth Script. A prompt alone is not enough, since GPT Image 2 received the same keep-list and still altered the plant (§ 3); the Rooted Renderer instead learns to follow it from its training pairs.

Pairs that reward real growth. A generator learns whatever its training pairs reward. Among the candidate pairs, however, long horizons are scarce, short ones are nearly solved by copying, and early stages appear mostly as sources and late ones as targets. We therefore fix a target share for each horizon range and, within each split, select candidate pairs within weekly quotas by a mixed-integer program that makes each stage appear about as often as a source as a target, with at least one pair per plant (Eq. (7) in Sec. D). From the candidates, we drop pairs whose change is not growth, such as backward stages or early yellowing, and draw the final pairs stratified by source and target stage; the fine-tuned baselines use them too.

Learning to follow the script. The Rooted Renderer should learn that following the script pays off, so we train it with the Hindsight Tutor’s scripts s⋆s^{\star}, whose stages and values match the target, and give it the scripts s^\hat{s} of PhenoClock only at test time; imperfect predicted scripts in training could teach it to ignore the script. We fit each triplet (xt,s⋆,xt+Δ)(x_t,s^{\star},x_{t+\Delta}) with the flow-matching objective of the base model [59, 60]:

Lren=Eτ,ϵ[∥vθ(ξτ,τ;E(xt),p⋆)−(ϵ−ξ0)∥22],(6)\mathcal{L}_{\mathrm{ren}} = \mathbb{E}_{\tau,\epsilon}\left[\left\|v_{\theta}(\xi_{\tau},\tau;\mathcal{E}(x_t),p^{\star})-(\epsilon-\xi_0)\right\|_2^2\right], \tag*{(6)}

where E\mathcal{E} is its VAE encoder, ξ0=E(xt+Δ)\xi_0=\mathcal{E}(x_{t+\Delta}), ξτ=(1−τ)ξ0+τϵ\xi_{\tau}=(1-\tau)\xi_0+\tau\epsilon with ϵ∼N(0,I)\epsilon\sim\mathcal{N}(0,I), and p⋆p^{\star} is the prompt built from (Δ,c,s⋆)(\Delta,c,s^{\star}); the text and VAE encoders stay frozen.

Experiments

We compare AgroWM with state-of-the-art image generators and domain-specific models on rice and maize, test it at long horizons, and assess its modules and the accuracy of its Growth Scripts.

Experimental Setup

Datasets. We validate AgroWM on two crops that grow very differently, with two public datasets of the Open Plant Image Archive (OPIA) [61]. WGSR [15] photographs 93 rice cultivars about once a week over the season; we keep its top view and one side (front) view with white backgrounds, labeled with the eight growth stages of § 5.2 and each cultivar’s heading date. WGSM [62] photographs 224 maize inbred lines across nine growth stages, from vegetative growth to physiological maturity, and we keep one side view. Each query pairs an earlier and a later photo of the same plant and view; rice pairs are 4 to 140 days apart (median 35). We split by plant: rice has 20,000 training queries from 252 plants and 2,000 validation queries from 79 held-out plants, 78 of whose cultivars also occur in training, and maize has 6,400 and 1,600 queries from 348 and 84 plants (Sec. D).

Baselines. The baselines fall into the two groups of Table 1. General-purpose generators are state-of-the-art image generation and editing models: the closed-source GPT Image 2 and GPT Image 2.5 and the open-source Qwen Image 2.1, Qwen-Image-Edit-2511 [63], FLUX.2 Klein, FLUX 3 Image Klein Base [64], and LANCE [65]. Each receives today’s photo and a prompt with the cultivar, view, and horizon, plus the Rooted Renderer’s keep-list, without a Growth Script (Sec. E). Domain-specific methods are built for this task. Copy returns today’s photo. LANCE FT, FLUX.2 Klein FT, Qwen Image 2.1 FT, and Agricrafter [12], a crop growth video generator, are fine-tuned on our data; method-specific training details are given in Sec. E.

Implementation details. We fine-tune PhenoClock from Qwen3.8-27B [66] with LoRA adapters [67] and the Rooted Renderer from Qwen Image 2.1 [63, 68], a generation and editing model with a 7B diffusion transformer, in all transformer weights; the Hindsight Tutor is GPT-6.1 Sol [69]. Both modules are trained for two epochs on the 20,000 training queries with a global batch of 16. At test time, the Rooted Renderer samples 20 denoising steps; the settings of all methods are in Sec. E.

Main Results

AgroWM outperforms all seven state-of-the-art generators on every identity, geometry, color, and timing score (Table 1), with Stage and Event R@1R@1 of 54.6% and 24.8%, against 34.8% and 10.8% for GPT Image 2.5 and 24.0% and 1.1% for Qwen Image 2.1. These generators change the plant too little or without control: Qwen Image 2.1 barely changes it, the other open models miss its area by 14 to 42 points, two to six times the error of the copy, and GPT Image 2, the best of them in Stage R@1R@1, reaches 35.3%. Fine-tuning on our pairs helps all three generators we fine-tune, raising Stage R@1R@1 from 24.5% to 46.9% for LANCE, from 25.2% to 39.3% for FLUX.2 Klein, and from 24.0% to 47.4% for Qwen Image 2.1, all above the 32.1% of Agricrafter. Adding Growth Scripts on top of the same backbone and pairs lifts AgroWM above every domain-specific model, by at least 7.2 and 2.3 points in Stage and Event R@1R@1. It also has the best identity and trait scores of all generated forecasts, except IoU, where LANCE FT and Agricrafter are slightly higher (0.342 and 0.329 against 0.327).

Generalization to Maize

Maize tests the framework on a crop with a single stalk, tassels, and ears. We train maize-specific models from the original pretrained backbones, adapting the stage scale and organ descriptions (Secs. D and E). AgroWM has the best overall means on the four reported metrics (Table 2: Stage R@1R@1 is 30.9%, against 30.1% for Qwen Image 2.1 FT and 21.1% for GPT Image 2, with area and height errors of 1.73 and 10.61 points. Its Stage R@1R@1 advantage over Qwen Image 2.1 FT is smaller than on rice.

MethodDINO↑Area↓Height↓Stage R@1↑
Copy0.5223.6233.020.0
GPT Image 20.7272.6316.6421.1
Qwen Image 2.1 FT0.7641.9711.1230.1
AgroWM0.7951.7310.6130.9

Table 2. Generalization to maize.

Long-Horizon Prediction

The further ahead a forecast looks, the less of the answer today’s photo shows, so Figure 4 splits the scores by horizon. One to four weeks ahead, the stage often stays the same, so the copy of today’s photo already reaches 45.8% Stage R@1R@1, and AgroWM and Qwen Image 2.1 FT are about equally accurate (Stage R@1R@1 62.8% and 62.4%, Event R@1R@1 27.3% and 27.8%).

As the horizon grows, all four methods of Figure 4 lose Stage R@1, AgroWM the least: from 1–4 to 9–20 weeks, it falls from 62.8% to 46.4%, against 62.4% to 35.7% for Qwen Image 2.1 FT and 49.2% to 22.4% for GPT Image 2. From five weeks on, AgroWM is best on all four scores of Figure 4. Its Stage R@1 lead over Qwen Image 2.1 FT is 0.4 points at 1–4 weeks but 13.9 and 10.7 points at 5–8 and 9–20 weeks, and at 9–20 weeks it also leads by 4.8 points in Event R@1 and lowers the area and yellow errors by 0.4 and 1.9 points. Since the two share backbone, pairs, and schedule and differ in whether they use Growth Scripts, reasoning about development helps most where today’s photo leaves the future open.

Ablation Study

We compare AgroWM with a script-free fine-tuning control, a script-only base generator, and a target-informed script reference in Table 3. Without PhenoClock (Qwen Image 2.1 FT, trained and run without Growth Scripts), Stage and Event R@1 drop by 7.2 and 2.3 points (95% CI 5.2–9.2 and 0.5–4.2, paired bootstrap over plants), and Stage R@1 by 10.7 points at 9–20 weeks (§ 6.4). Without the Rooted Renderer, the base Qwen Image 2.1 receives the Growth Scripts of PhenoClock in its prompt but neither the source photo nor fine-tuning: Stage and Event R@1 fall to 24.8% and 12.3%, the area error rises to 8.22 points, and DINO drops to 0.565, so a good script is not enough without a renderer that keeps the plant. With the Hindsight Tutor’s target-informed scripts, the same Rooted Renderer reaches 58.9% and 28.6% (Sec. 5.6).

VariantStage R@1↑Event R@1↑Area↓DINO↑
AgroWM (full)54.624.83.510.860
w/o PhenoClock47.422.53.730.834
w/o Rooted Renderer24.812.38.220.565
Hindsight Tutor (Upper Bound)58.928.62.920.874

Table 3. Ablation studies on rice. R@1 in %, area error in points.

How Accurate Is the Growth Script?

Errors in the Growth Script carry over to the forecast, so we check the scripts of PhenoClock against the recorded stages and measured traits of the real future (Table 4). PhenoClock names today’s stage correctly in 89.9% of the queries and the future stage in 78.9%, and 99.4% of its future stages are at most one stage off, so advancing the stage is harder than reading it. Its target areas are off by 4.81, 2.16, and 8.24 points for height, area, and yellow, and 1,999 of its 2,000 scripts pass the strict format check. These errors cost 4.3 points of Stage R@1: with the Rooted Renderer fixed, the Hindsight Tutor’s scripts reach 58.9% and those of PhenoClock 54.6% (95% CI of the gap 2.2–6.3; Table 3), so better scripts would give better forecasts.

CurrentFutureFuture ±1
Stage accuracy (%) ↑89.9078.9099.40
HeightAreaYellow
Target MAE (pp) ↓4.812.168.24
Change-direction Macro-F1 ↑0.6590.7080.556
Semantic similarity (cosine) ↑0.842

Table 4. Growth Script accuracy on the rice validation split, against the recorded stages and measured slots of the real future (height: 1,000 front views). ±1: within one stage. Similarity: mean cosine similarity of BGE [70] embeddings to the Hindsight Tutor’s 18,000 slot descriptions.

Conclusion and Future Work

We asked whether world models can forecast the slow change of living things, and found that two state-of-the-art image generators either barely change a rice plant or change it at the wrong pace and stage, while pixel scores favor the unchanged plant. AgroWM narrows this gap by reasoning before rendering: it first forecasts a Growth Script of the plant’s development, learned from scripts written in hindsight, and then edits the photo to match. Reasoning adds 7.2 points of Stage R@1 over the same backbone without Growth Scripts, most beyond four weeks, and puts AgroWM ahead of state-of-the-art generators and domain-specific models on both rice and maize, two crops with different body plans and stage scales. Because each forecast is a photo of the same plant, we can check its identity, size, color, and developmental timing against the plant’s own later photos. The rice results come from potted plants on one greenhouse platform, with the background removed, mostly of cultivars seen in training, and from pairs filtered for growth. Future work will condition forecasts on weather and management records and test them on new cultivars, more species, and field images.

Supplementary Material

More Cases of the Preliminary Analysis

Figure 2 in § 3 shows one case per failure, chosen so that AgroWM is best on every trait error and retrieval rank shown. In its top-view rows (a) and (c), PSNR and SSIM also prefer Qwen Image 2.1 or GPT Image 2, and this holds in only 3 of the 64 cases; in its front-view row (b), one of only 2 of the 32 front views in which every plant and timing score prefers AgroWM, SSIM prefers Qwen Image 2.1 while PSNR prefers AgroWM. Figure S1 shows two cases per failure, chosen for the failure alone and covering both views, including the front-view failures that § 4 reports as statistics. Figure S2 shows all three, rows (a) and (c) of Figure 2 and one front view, with the score labels in full and the share of cases in which each score prefers no growth, grouped as in Table S1.

Figure S2. What each score sees. The three cases in which PSNR and SSIM prefer Qwen Image 2.1 or GPT Image 2 while AgroWM is best or tied on every plant and timing score; (a) and (c) are rows (a) and (c) of Figure 2.

The cases of Figure S1 add details that § 4 gives only as statistics. GPT Image 2 paces its edits only loosely by the interval: after three weeks, it adds panicles to a plant that is still jointing (Figure S1), whereas after eleven weeks, it keeps a young tillering plant at 6× its area, against 65× for the real plant (Figure S1). In front views, it turns a plant that is still jointing 72% yellow or brown, against 22% in reality (Figure S1), and draws a plant taller, reaching 87% of the image height where the real plant stays at 58% (Figure S1). Ripening plants stay green in both models (Figure S1), most of all in top views, where the real plants are on average 63% yellow or brown, against 9% for GPT Image 2. Qwen Image 2.1 keeps the plant mostly by not changing it, but in 4 of the 64 cases it replaces the plant with a single culm or a field texture.

Analysis of Each Score

This section expands § 4.1. We compare two forecasts with little or no growth, the exact copy of today’s photo and Qwen Image 2.1, which barely changes it, with that of AgroWM, which sees only today’s photo, the horizon, and the conditions: Table S1 counts how often each score prefers no growth over the 2,000 validation queries, and the 64 cases below show why. All scores are computed by one evaluator on the white-background images.

Background agreement can dominate pixel scores. Over the 64 cases, the copy has a mean PSNR of 14.9 dB and a mean SSIM of 0.81, against 14.7 dB and 0.78 for Qwen Image 2.1 and 14.2 dB and 0.75 for GPT Image 2. Plants cover on average 14% of a real future photo in the 64 cases, and 17% over the validation set; the rest is white background. PSNR and SSIM therefore weigh mostly pixels that a forecast leaves untouched when it does not grow, and they penalize a leaf drawn slightly off twice, where it is and where it should be, but a missing leaf only once. SSIM prefers the copy to AgroWM in 63 of the 64 cases, and PSNR in 38. In Figure S2, the only forecast with the right size and stage receives the lowest PSNR and SSIM of the three, 1.3 dB below Qwen Image 2.1. Over the 2,000 validation queries, Qwen Image 2.1, which barely changes the plant, has the highest PSNR of all deployable models in Table 1, only the copy has a higher SSIM, and both are above AgroWM.

Global features are not anchored in time. Cosine similarity of DINOv2 features [39] between the forecast and the real future depends little on the background and prefers the copy to AgroWM in only 8 of 63 cases. Its value, however, depends on the horizon as much as on the forecast: the copy scores 0.74 at 1–4 weeks and 0.41 at 9–20 weeks, so the same value can mean a good long-horizon forecast or no change at a short horizon. On the validation set, the copy reaches 0.702, higher than five of the seven general-purpose generators and below only GPT Image 2 and GPT Image 2.5 (0.709 and 0.721). A single similarity also cannot say in which direction a forecast is wrong, whether too early, too late, or not the same plant.

Plant traits measure growth, but no single trait measures time. Measured on the plant pixels alone, the errors in visible area, vertical extent, and yellow–brown share see the growth that pixel scores miss: in Figure S2 Qwen Image 2.1 is 20.4 and 40.2 points off in area and extent, against 0.5 and 0.0 for AgroWM. Each trait, however, sees only one side of development. A forecast of the right size can be at the wrong stage: GPT Image 2 has a smaller area error than the copy and Qwen Image 2.1 in 52 of the 64 cases (5.4 points on average over the 64 cases, against 10.0 for the copy), yet 33 of these 52 forecasts miss the target stage under Stage R@1. Size cannot tell a green plant from a ripening one: in a top-view case twelve weeks ahead, GPT Image 2 misses the real area by only 3.6 points, but its plant is green where the Culm: Culms extend upward from the same basal crown as the established tillers progress from short basal growth toward jointing-stage elongation. Metric = Height; Source = 23.44%; Target = 62.39%; Change = +38.95 pp.

real one is 94% yellow and brown. Color, in turn, is informative only when color changes: over all 64 cases Qwen Image 2.1 beats AgroWM on yellow–brown error in 18, because most real futures are still green, but in only 4 of the 22 cases that reach grain filling or ripening, such as Figure S2, where it is 17.4 points off and AgroWM 2.4. In front views, every plant reaches the bottom edge (all 64 source and target images), so the vertical extent equals the culm slot of the Growth Script there.

Timing scores and no growth. The 36 of 2,000 queries whose plant has fewer than five photos in the view are skipped. Since the gallery contains today’s photo, the copy has an Event R@1 of zero, and its Stage R@1 is one only where the stage does not change, which is then the right answer. Accordingly, Event R@1 never prefers the copy to AgroWM, and Stage R@1 does so in 3 of 63 cases, all with an unchanged stage.

Slots of a Growth Script

Table S2 lists the slots of a Growth Script (§ 5.2) and how each is measured.

Table 4 in Sec. 5.6 scores the Growth Scripts of PhenoClock against the recorded stages and measured slots of the real future. Most of its future-stage errors are off by one stage, its direction of change is most often right for area and least for yellow, its descriptions read much like the Hindsight Tutor’s, and the one script that fails the strict format check states a height change that does not match its values and is passed to the Rooted Renderer as written.

Change-direction Macro-F1. We classify each predicted and reference source-to-target change as decrease (d<−1d < -1 pp), stable (−1≤d≤1-1 \le d \le1 pp), or increase (d>1d > 1 pp), where dd is the change in the corresponding image proxy. For each of height, area, and yellow–brown share, Macro-F1 is the unweighted mean of the F1 scores of these three classes.

Description similarity. We align the predicted and Hindsight Tutor descriptions by view, section (Growth plan or Target phenotype), and slot, and remove the numeric evidence suffixes before comparison. The 2,000 validation queries contain 18,000 reference descriptions across these two sections: five slots per front view and four per top view. We encode the descriptions with BAAI/bge-small-en-v1.5 [70], use L2-normalized CLS embeddings, and report the mean cosine similarity over matched descriptions. This score measures agreement with the Hindsight Tutor’s wording and meaning; the stage and numeric scores assess factual accuracy.

Datasets, Cleaning, and Training Pairs

WGSR [15], a public dataset of the Open Plant Image Archive [61], holds 16,758 RGB photos of 93 rice accessions, taken from two sides and from the top about once a week from 42 to 182 days after sowing, on a LemnaTec phenotyping platform in a greenhouse in Beijing. We keep the top view and one side view, called the front view, and generate white-background versions with Qwen Image 2.1 [68], whose prompt, rewritten per photo by its prompt enhancer, asks to keep the plant unchanged and to replace everything else with white, so that methods and scores see only the plant. Some outputs keep remnants of the pot or rails or shift the color of young leaves; the color-coverage gate and the pair gates below remove part of them. Each photo carries one of the eight growth stages of § 5.2, and each accession its heading date. To test on unseen plants, one random eligible plant per accession goes to validation before pairs are filtered and selected, and all photos of a plant stay in one split. Stage-stratified sampling yields 10,000 training and 1,000 validation pairs, each in both views: 20,000 training queries from 252 plants of 85 accessions and 2,000 validation queries from 79 plants, one from each of 79 accessions, 78 of which also occur in training. These queries use 7,540 photos of 331 plants from 86 accessions.

Before pairing, we drop the photos recorded as abnormal growth (three accessions and 57 single photos), all photos from the day a plant first received an artificial support, and a few others by hand. Within each split, the candidate selection of § 5.3 solves

min⁡y∈{0,1}P∑z∣Nzsrc−Nztgt∣+λ∑z(∣Nzsrc−Nˉz∣+∣Nztgt−Nˉz∣),(7)\min_{\mathbf{y}\in\{0,1\}^{P}} \sum_z \left|N^{\mathrm{src}}_z-N^{\mathrm{tgt}}_z\right|+\lambda\sum_z\left(\left|N^{\mathrm{src}}_z-\bar{N}_z\right|+\left|N^{\mathrm{tgt}}_z-\bar{N}_z\right|\right), \tag*{(7)}

subject to weekly quotas and at least one pair per plant, where y\mathbf{y} selects among PP candidates, NzsrcN^{\mathrm{src}}_z and NztgtN^{\mathrm{tgt}}_z count selected sources and targets in stage zz, Nˉz\bar{N}_z is the count under the pool’s stage proportions, and λ\lambda weighs the second term. It targets 40%, 35%, and 25% of the pairs at 1–4, 5–8, and 9–20 weeks; after the gates and the stratified sampling, 46%, 33%, and 21% of the validation queries fall in these ranges, so 54.4% are beyond four weeks. The gates drop pairs whose change is not growth: photos at most three days apart; backward stages or panicles; early yellowing, that is, a yellow–brown share of at least 0.35 in tillering or jointing, or a rise of at least 0.20 from these stages within two stages; a front-height loss of at least 5 points toward a target before yellow ripening; and color coverage below 70%. We attribute such changes to background removal, lighting, or leaning plants rather than growth; a pair that kept them would teach the Rooted Renderer to draw the artifact and the Hindsight Tutor to explain it.

WGSM [62], also available through OPIA, contains images of 224 maize inbred lines acquired on a high-throughput phenotyping platform across nine growth stages, which cover vegetative growth, tasseling, silking, grain filling, and physiological maturity. We retain one side view and construct strictly forward pairs within each plant; the 768×1,270768 \times1{,}270 source images are padded by 10 pixels at the top to a 768×1,280768 \times1{,}280 canvas, then processed to a white background with connected-component cleanup. The resulting queries cover 432 plants, split by plant into 348 training and 84 held-out validation plants, with 6,400 and 1,600 queries (§ 6.3). The protocol maps stages S1–S9 to V3, V5, V7, V10, V14, VT, R1, R3, and R6, respectively, and uses the unrounded interval in days; the Planner receives the plant ID, source image, and interval, but not the recorded query stages or future image.

Implementation Details

For rice, photos are 672×896672 \times896 pixels (width×\timesheight) in the front view and 896×672896 \times672 in the top view. The candidate selection of Eq. (7) uses λ=0.02\lambda= 0.02; the Hindsight Tutor runs with reasoning effort xhighx_{\mathrm{high}}, and all 22,000 queries end with a Hindsight Tutor script that passes the validator. Table S3 summarizes the training recipes for the learned methods reported in the paper.

Model / armInitialization and trainable componentsTraining dataRecipe
Rice PlannerQwen3.8-27B; LoRA r=32r = 32, α=64\alpha= 64; all linear layers; vision tower and aligner frozen20,000 queries; global batch 162 epochs, 2,500 updates; LR 10−410^{-4}, 3% warmup, cosine to 10−610^{-6}; grouped loss 0.1/0.4/0.5
Rice Simple Executor (Qwen Image 2.1 FT)Original Qwen Image 2.1; full DiT; text and VAE encoders frozen; no Growth Script20,000 queries; global batch 162 epochs, 2,500 updates; peak LR 10−510^{-5}; 10% warmup, 60% stable, 30% cosine to 10−610^{-6}
Rice CoT Executor (Rooted Renderer)Original Qwen Image 2.1; full DiT; text and VAE encoders frozen; teacher Growth Scripts in training20,000 queries; global batch 162 epochs, 2,500 updates; peak LR 10−510^{-5}; 10% warmup, 60% stable, 30% cosine to 10−610^{-6}; 8 GPUs
Rice FLUX.2 Klein FT / LANCE FTCorresponding pretrained image model; fine-tuned image-generation backbone; no Planner or Growth Script20,000 queries; global batch 162 epochs, 2,500 updates; LR 10−510^{-5}; 10% warmup, 60% stable, 30% cosine to 10−610^{-6}; 8 GPUs
AgricrafterDynamiCrafter_512 initialization; UNet and image projector trained2,108 nine-observation windows from 178 rice training plants; global batch 46,000 updates; white letterbox canvas 320×512320 \times512
Maize PlannerQwen3.8-27B; LoRA r=32r = 32, α=64\alpha= 64; all linear layers; vision tower and aligner frozen6,400 queries; global batch 162 epochs, 800 updates; LR 10−410^{-4}, 3% warmup, cosine to 10−610^{-6}; grouped loss 0.1/0.4/0.5
Maize Simple / CoT ExecutorsSeparately initialized Qwen Image 2.1 full DiTs; text and VAE encoders frozen; no script / teacher script in training6,400 queries per arm; global batch 162 epochs, 800 updates; peak LR 10−510^{-5}; 10% warmup, 60% stable, 30% cosine to 10−610^{-6}

Table S3. Training configurations for paper methods. All local Planner and Executor runs use bfloat16. The image-model schedules use a peak learning rate followed by the listed warmup, stable, and cosine-decay fractions.

At test time, the Rooted Renderer and Qwen Image 2.1 FT sample 20 denoising steps with a guidance scale of 1 in bf16, at the source size. Among the open general-purpose baselines, Qwen Image 2.1, Qwen-Image-Edit-2511, and the 9B FLUX.2 Klein base model use 20 denoising steps in bf16, with true classifier-free guidance 1.0, true guidance 4.0, and guidance 4.0, respectively; Qwen Image 2.1 outputs are composited on white. The 3B LANCE uses its official image-editing pipeline with 20 steps. GPT Image 2 is called through its image-editing API with automatic quality, and its outputs are resized to the source size with Lanczos filtering. GPT Image 2.5 and FLUX 3 Image Klein Base are included as evaluated in the main table; settings not present in the run records are not inferred here. The rice image generators receive the keep-list of the Rooted Renderer’s prompt and the horizon in rounded whole weeks. All forecasts are scored at the source size. For FLUX 3 Image Klein Base, only 998 queries have a valid height and 1,996 a valid density and yellow share.

Baseline Experiments

This section describes the baselines reported in the paper’s main comparison and maize comparison (Table 2). The copy returns the source image unchanged. The general-purpose image generators receive the source image and the task prompt described in § 6.1, including the accession, view, horizon, and the Rooted Renderer keep-list, without a predicted Growth Script. The open-model inference settings and output-size handling are given above; generated images are evaluated at the source dimensions.

The rice Qwen Image 2.1 FT, FLUX.2 Klein FT, and LANCE FT baselines are trained on the same rice training split without a Planner or Growth Script; their training recipes are listed in Table S3. Qwen Image 2.1 FT is the Simple Executor control for the Rooted Renderer. The GPT Image API baselines and the copy do not use these training updates.

Agricrafter is a video-generation baseline adapted to WGSR, rather than an image-pair fine-tune; its window construction and training recipe are listed in Table S3. At inference it predicts nine frames with DDIM for 50 steps and recursively advances by at most eight observation steps; the target frame is selected by the recorded observation-index difference, not by a requested number of days or weeks.

The maize comparison contains the source copy, GPT Image 2, a Simple Qwen Image 2.1 Executor, and AgroWM. The Planner, Simple Executor, and CoT Executor are each trained separately from their original pretrained model on the maize split; no rice-fine-tuned checkpoint is transferred. Their configurations are listed in Table S3. The CoT Executor is trained on teacher scripts but receives only the source-only Planner prediction at inference; GPT Image 2 uses the same source image and Simple task prompt, while the copy is unchanged.

Planner Output Templates

Each response contains one <GROWTH_COT> block with three sections in order: Stage reasoning, Growth plan, and Target phenotype. Figure S3 shows the crop- and view-specific templates with their ordered fields and value syntax; current/future stages and numeric targets are predictions, not labels supplied with the query.

CoT output templates for rice and maize

Figure S3. CoT output templates for rice and maize. Each column shows the complete three-section wrapper and ordered fields. Braces denote generated descriptions or values rather than literal output tokens; each field occupies one logical line and wraps here for readability. Growth plan describes the transition and gives proxy evidence, while Target phenotype describes the visible endpoint qualitatively. Image-proxy definitions are in Table S2.

For rice, the Planner receives the source image, accession, view, and rounded interval in weeks. Growth plan fields may include an optional Guardrails: clause after the change description; unavailable metric evidence is omitted. Rice numeric proxies use two decimal places, while Panicle uses visible/not visible and a signed visibility change. The top-view template uses Canopy density and omits the front-view Culm/Height field.

For maize, the query supplies the source image, plant ID, side view, and exact interval in days. The four numeric fields use one decimal place; Tassel and Ear contain only qualitative visibility descriptions, and hidden organs are not reconstructed. When the Culm change is within 2% of the source proxy, its numeric triplet may be replaced by the exact wording Stable within the registered two-percent proxy tolerance; Source/Target/Change suppressed. The maize Planner’s system prompt includes a general guide listing all nine stage IDs S1–S9, their corresponding growth stages, and visible descriptions; it does not identify the current or future stage of the query plant. The dataset protocol maps S1–S9 to V3, V5, V7, V10, V14, VT, R1, R3, and R6 at nominal 25, 32, 50, 59, 70, 78, 85, 91, and 113 days after sowing; these protocol reference days do not imply that every plant was imaged on those dates.

Evaluation Coverage

The table reports the valid query count for each score in the paper; unavailable phenotype measurements remain missing and are not counted as zero error. Rice has 2,000 validation queries in total. Height applies to front views only, while DINOv2 and stage/event retrieval require an eligible same-plant, same-view photo series.

Dataset / methodValid queries by score
Rice, most methodsN=2,000N = 2{,}000: area, density, yellow errors; IoU, PSNR, SSIM; N=1,000N = 1{,}000: height (front); N=1,964N = 1{,}964: DINOv2, Stage R@1, Event R@1
Rice, FLUX 3N=998N = 998: height; N=1,996N = 1{,}996: density and yellow
Maize, all methodsN=1,600N = 1{,}600: DINOv2, Stage R@1, pixel scores
Maize, copy / GPT Image 2N=1,600N = 1{,}600: area, culm-proxy height, density, yellow errors
Maize, Simple / AgroWMN=1,596N = 1{,}596 each: area, culm-proxy height, density; N=1,592/1,591N = 1{,}592/1{,}591: yellow error

Table S4. Evaluation coverage in the paper. Counts are valid queries over the indicated dataset.

Maize phenotype counts are lower for trained image models when the foreground is empty or does not contain enough valid color pixels; the underlying predictions remain in the evaluation set. The maize evaluation uses one view, and its gallery has one image per plant, view, and stage, so Stage and Event retrieval hits coincide; the paper reports Stage R@1.

References

  1. [1]Tim Brooks, Bill Peebles, et al. Video generation models as world simulators. OpenAI technical report, 2024.
  2. [2]Anthony Hu, Lloyd Russell, Hudson Yeo, Zak Murez, George Fedoseev, Alex Kendall, Jamie Shotton, and Gianluca Corrado. GAIA-1: A generative world model for autonomous driving. arXiv preprint arXiv:2309.17080, 2023.arxiv.org/abs/2309.17080
  3. [3]Sherry Yang, Yilun Du, Kamyar Ghassempour, Jonathan Tompson, Leslie Pack Kaelbling, Dale Schuurmans, and Pieter Abbeel. Learning interactive real-world simulators. In ICLR, 2024.arxiv.org/abs/2310.06114
  4. [4]Sumithra Muthayya, Jonathan D. Sugimoto, Scott Montgomery, and Glen F. Maberly. An overview of global rice production, supply, trade, and consumption. Annals of the New York Academy of Sciences, 1324(1):7–14, 2014.DOI
  5. [5]Shenghai Yuan, Jinfa Huang, Yujun Shi, et al. MagicTime: Time-lapse video generation models as metamorphic simulators. IEEE TPAMI, 2025.DOI
  6. [6]Shenghai Yuan, Jinfa Huang, Yongqi Xu, et al. ChronoMagic-Bench: A benchmark for metamorphic evaluation of text-to-time-lapse video generation. In NeurIPS Datasets and Benchmarks Track, 2024.arxiv.org/abs/2406.18522
  7. [7]Xiaojuan Wang, Boyang Zhou, Brian Curless, Ira Kemelmacher-Shlizerman, Aleksander Holynski, and Steven M. Seitz. Generative inbetweening: Adapting image-to-video models for keyframe interpolation. In ICLR, 2025.arxiv.org/abs/2408.15239
  8. [8]Weihan Luo, Lily Goli, Sherwin Bahmani, Felix Taubner, Andrea Tagliasacchi, and David B. Lindell. Grow with the flow: 4D reconstruction of growing plants with Gaussian flow fields. arXiv preprint arXiv:2602.08958, 2026.arxiv.org/abs/2602.08958
  9. [9]Lukas Drees, Laura Verena Junker-Froh, Jana Kierdorf, and Ribana Roscher. Temporal prediction and evaluation of Brassica growth in the field using conditional generative adversarial networks. arXiv preprint arXiv:2105.07789, 2021.DOI
  10. [10]Lukas Drees, Dereje T. Demie, Madhuri R. Paul, Johannes Leonhardt, Sabine J. Seidel, Thomas F. Döring, and Ribana Roscher. Data-driven crop growth simulation on time-varying generated images using multi-conditional generative adversarial networks. Plant Methods, 20:93, 2024.DOI
  11. [11]Xinyi Wang et al. GAN-based image prediction of maize growth across varieties and developmental stages. Plant Methods, 21:110, 2025.DOI
  12. [12]Lining Wang, Zhan Zhang, and Hongxun Yao. Agricrafter: Full-growth-cycle crop video generation via diffusion models. Computers and Electronics in Agriculture, 239:110946, 2025.DOI
  13. [13]B. A. M. Bouman, M. J. Kropff, T. P. Tuong, M. C. S. Wopereis, H. F. M. ten Berge, and H. H. van Laar. ORYZA2000: Modeling Lowland Rice. International Rice Research Institute, Los Baños, Philippines, 2001.DOI
  14. [14]James W. Jones, Gerrit Hoogenboom, Cheryl H. Porter, et al. The DSSAT cropping system model. European Journal of Agronomy, 18(3–4):235–265, 2003.DOI
  15. [15]Zhixin Tang, Zhuo Chen, Yuan Gao, Ruxian Xue, Zedong Geng, Qingyun Bu, Yanyan Wang, Xiaoqian Chen, Yuqiang Jiang, Fan Chen, Wanneng Yang, and Weijuan Hu. A strategy for the acquisition and analysis of image-based phenome in rice during the whole growth period. Plant Phenomics, 5:0058, 2023.DOI
  16. [16]David Ha and Jürgen Schmidhuber. Recurrent world models facilitate policy evolution. In NeurIPS, 2018.arxiv.org/abs/1809.01999
  17. [17]Danijar Hafner, Jurgen Pasukonis, Jimmy Ba, and Timothy Lillicrap. Mastering diverse domains through world models. arXiv preprint arXiv:2301.04104, 2023.arxiv.org/abs/2301.04104
  18. [18]NVIDIA. Cosmos world foundation model platform for physical AI. arXiv preprint arXiv:2501.03575, 2025.
  19. [19]Wan Team. Wan: Open and advanced large-scale video generative models. arXiv preprint arXiv:2503.20314, 2025.arxiv.org/abs/2503.20314
  20. [20]Jake Bruce, Michael Dennis, Ashley Edwards, et al. Genie: Generative interactive environments. In ICML, 2024.arxiv.org/abs/2402.15391
  21. [21]Dani Valevski, Yaniv Leviathan, Moab Arar, and Shlomi Fruchter. Diffusion models are real-time game engines. arXiv preprint arXiv:2408.14837, 2024.arxiv.org/abs/2408.14837
  22. [22]Mido Assran et al. V-JEPA 2: Self-supervised video models enable understanding, prediction and planning. arXiv preprint arXiv:2506.09985, 2025.arxiv.org/abs/2506.09985
  23. [23]Alexander Neitz, Giambattista Parascondolo, Stefan Bauer, and Bernhard Schölkopf. Adaptive skip intervals: Temporal abstraction for recurrent dynamical models. In NeurIPS, 2018.arxiv.org/abs/1808.04768
  24. [24]Karol Gregor, George Papamakarios, Frederic Besse, Lars Buesing, and Théophane Weber. Temporal difference variational auto-encoder. In ICLR, 2019.arxiv.org/abs/1806.03107
  25. [25]Dinesh Jayaraman, Frederik Ebert, Alexei A. Efros, and Sergey Levine. Time-agnostic prediction: Predicting predictable video frames. In ICLR, 2019.
  26. [26]Vaibhav Saxena, Jimmy Ba, and Danijar Hafner. Clockwork variational autoencoders. In NeurIPS, 2021.arxiv.org/abs/2102.09532
  27. [27]Jesse Farebrother, Matteo Pirotta, Andrea Tirinzoni, Marc G. Bellemare, Alessandro Lazaric, and Ahmed Touati. Compositional planning with jumpy world models. arXiv preprint arXiv:2602.19634, 2026.arxiv.org/abs/2602.19634
  28. [28]Remi Lam, Alvaro Sanchez-Gonzalez, Matthew Willson, et al. Learning skillful medium-range global weather forecasting. Science, 382(6677):1416–1421, 2023.DOI
  29. [29]Kaifeng Bi, Lingxi Xie, Hengheng Zhang, Xin Chen, Xiaotao Gu, and Qi Tian. Accurate medium-range global weather forecasting with 3D neural networks. Nature, 619(7970):533–538, 2023.DOI
  30. [30]Hritik Bansal, Zongyu Lin, Tianyi Xie, et al. VideoPhy: Evaluating physical commonsense for video generation. arXiv preprint arXiv:2406.03520, 2024.arxiv.org/abs/2406.03520
  31. [31]Bingyi Kang, Yang Yue, Rui Lu, et al. How far is video generation from world model: A physical law perspective. arXiv preprint arXiv:2411.02385, 2024.arxiv.org/abs/2411.02385
  32. [32]Saman Motamed, Laura Culp, Kevin Swersky, Priyank Jain, and Robert Geirhos. Do generative video models understand physical principles? In WACV, 2026.arxiv.org/abs/2501.09038
  33. [33]Mingyang Wu, Siyuan Yang, Jiongze Yu, Pardis Taghavi, Fangzhou Lin, and Zhengzhong Tu. The pulse of motion: Measuring physical frame rate from visual dynamics. arXiv preprint arXiv:2603.14375, 2026.arxiv.org/abs/2603.14375
  34. [34]Yipin Zhou and Tamara L. Berg. Learning temporal transformations from time-lapse videos. In ECCV, 2016.arxiv.org/abs/1608.07724
  35. [35]Wei Xiong, Wenhan Luo, Lin Ma, Wei Liu, and Jiebo Luo. Learning to generate time-lapse videos using multi-stage dynamic generative adversarial networks. In CVPR, 2018.arxiv.org/abs/1709.07592
  36. [36]Roy Or-El, Soumyadip Sengupta, Ohad Fried, Eli Shechtman, and Ira Kemelmacher-Shlizerman. Lifespan age transformation synthesis. In ECCV, 2020.
  37. [37]Zhou Wang, Alan C. Bovik, Hamid R. Sheikh, and Eero P. Simoncelli. Image quality assessment: From error visibility to structural similarity. IEEE TIP, 13(4):600–612, 2004.pubmed.ncbi.nlm.nih.gov/15376593
  38. [38]Richard Zhang, Phillip Isola, Alexei A. Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In CVPR, 2018.arxiv.org/abs/1801.03924
  39. [39]Maxime Oquab, Timothée Darcet, Théo Moutakanni, et al. DINOv2: Learning robust visual features without supervision. Transactions on Machine Learning Research, 2024.
  40. [40]Yochai Blau and Tomer Michaeli. The perception-distortion tradeoff. In CVPR, 2018.DOI
  41. [41]Emily Denton and Rob Fergus. Stochastic video generation with a learned prior. In ICML, 2018.arxiv.org/abs/1802.07687
  42. [42]Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. GANs trained by a two time-scale update rule converge to a local Nash equilibrium. In NeurIPS, 2017.
  43. [43]Tsu-Jui Fu, Wenze Hu, Xianzhi Du, William Yang Wang, Yinfei Yang, and Zhe Gan. Guiding instruction-based image editing via multimodal large language models. In ICLR, 2024.arxiv.org/abs/2309.17102
  44. [44]Yuzhou Huang, Liangbin Xie, Xintao Wang, et al. SmartEdit: Exploring complex instruction-based image editing with multimodal large language models. In CVPR, 2024.arxiv.org/abs/2312.06739
  45. [45]Rongyao Fang, Chengqi Duan, Kun Wang, Linjiang Huang, Hao Li, Shilin Yan, Hao Tian, Xingyu Zeng, Rui Zhao, Yifeng Dai, Xihui Liu, and Hongsheng Li. GoT: Unleashing reasoning capability of multimodal large language model for visual generation and editing. arXiv preprint arXiv:2503.10639, 2025.arxiv.org/abs/2503.10639
  46. [46]Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc V. Le, and Denny Zhou. Chain-of-thought prompting elicits reasoning in large language models. In NeurIPS, 2022.arxiv.org/abs/2201.11903
  47. [47]Ziming Xu, Shuang Liang, Ruobing Han, Ziqiao Xi, Mingxing Rao, Kun Zhou, Zijun Zhang, Yuchen Yan, Yufan Wei, Junbo Huang, Yifei Shao, Fang Nan, and Biwei Huang. CausalWM: Causal chain-of-thought reasoning for embodied world model. arXiv preprint arXiv:2609.23184, 2026.arxiv.org/abs/2609.23184
  48. [48]Vladimir Vapnik and Akshay Vashist. A new learning paradigm: Learning using privileged information. Neural Networks, 22(5–6):544–557, 2009.DOI
  49. [49]Dian Chen, Brady Zhou, Vladlen Koltun, and Philipp Krähenbühl. Learning by cheating. In CoRL, 2019.arxiv.org/abs/1912.12294
  50. [50]Eric Zelikman, Yuhuai Wu, Jesse Mu, and Noah D. Goodman. STaR: Bootstrapping reasoning with reasoning. In NeurIPS, 2022.arxiv.org/abs/2203.14465
  51. [51]Dean P. Holzworth, Neil I. Huth, Peter G. deVoil, et al. APSIM – evolution towards a new generation of agricultural systems simulation. Environmental Modelling & Software, 62:327–350, 2014.
  52. [52]Allard de Wit, Hendrik Boogaard, Davide Fumagalli, et al. 25 years of the WOFOST cropping systems model. Agricultural Systems, 168:154–167, 2019.DOI
  53. [53]Przemyslaw Prusinkiewicz and Aristid Lindenmayer. The Algorithmic Beauty of Plants. Springer, New York, 1990.DOI
  54. [54]J. Vos, J. B. Evers, G. H. Buck-Sorlin, B. Andrieu, M. C.helle, and P. H. B. de Visser. Functional–structural plant modelling: A new versatile tool in crop science. Journal of Experimental Botany, 61(8):2101–2115, 2010.DOI
  55. [55]Hanno Scharr, Massimo Minervini, Andrew P. French, et al. Leaf segmentation in plant phenotyping: A collation study. Machine Vision and Applications, 27(4):585–606, 2016.DOI
  56. [56]Wanneng Yang, Zilong Guo, Chenglong Huang, et al. Combining high-throughput phenotyping and genome-wide association studies to reveal natural genetic variation in rice. Nature Communications, 5:5087, 2014.DOI
  57. [57]Ruchi Bhatt, Shreya Bansal, Amanpreet Chander, Rupinder Kaur, M.alya Singh, Mohan Kankanhalli, Abdulmotaleb El Saddik, and Mukesh Kumar Saini. GroMo: Plant growth modeling with multiview images. arXiv preprint arXiv:2503.06608, 2025.arxiv.org/abs/2503.06608
  58. [58]International Rice Research Institute, Manila, Philippines. Standard Evaluation System for Rice (SES), 5th edition, 2013.
  59. [59]Yaron Lipman, Ricky T. Q. Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow matching for generative modeling. In ICLR, 2023.arxiv.org/abs/2210.02747
  60. [60]Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow. In ICLR, 2023.arxiv.org/abs/2209.03003
  61. [61]Yongrong Cao, Dongmei Tian, Zhixin Tang, Xiaonan Liu, Weijuan Hu, Zhang Zhang, and Shuhui Song. OPIA: An open archive of plant images and related phenotypic traits. Nucleic Acids Research, 52(D1):D1530–D1537, 2024.DOI
  62. [62]Weixuan Wang, Weijun Guo, Liang Le, Jia Yu, Yue Wu, Dongwei Li, Yifan Wang, Huan Wang, Xiaoduo Lu, Hong Qiao, Xiaofeng Gu, Jian Tian, Chunyi Zhang, and Li Pu. Integration of high-throughput phenotyping, GWAS, and predictive models reveals the genetic architecture of plant height in maize. Molecular Plant, 16(2):354–373, 2023.DOI
  63. [63]Chenfei Wu, Jiahao Li, Jingren Zhou, et al. Qwen-Image technical report. arXiv preprint arXiv:2508.02324, 2025.arxiv.org/abs/2508.02324
  64. [64]Black Forest Labs. FLUX 3 Image Klein Base. Hugging Face model card, 2026.
  65. [65]Fengyi Fu, Mengqi Huang, Shaojin Wu, Yunsheng Jiang, Yufei Huo, Hao Li, Yinghang Song, Fei Ding, Jianzhu Guo, Qian He, Zheren Fu, Zhendong Mao, and Yongdong Zhang. Lance: Unified multimodal modeling by multi-task synergy. arXiv preprint arXiv:2605.18768, 2026.
  66. [66]Qwen Team. Qwen3.8-27B. Hugging Face model card, 2026.
  67. [67]Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. LoRA: Low-rank adaptation of large language models. In ICLR, 2022.arxiv.org/abs/2106.09685
  68. [68]Qwen Team. Qwen-Image-2.1. GitHub repository, 2026.
  69. [69]OpenAI. Introducing GPT-6.1 Sol. OpenAI blog, 2026.
  70. [70]Shitao Xiao, Zheng Liu, Peitian Zhang, Niklas Muennighoff, Defu Lian, and Jian-Yun Nie. C-Pack: Packed resources for general Chinese embeddings. In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval, 2024.arxiv.org/abs/2309.07597

Paper details

vlnc:W_dwe4grwf
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Computer Science

Not recorded for this paper

View statement
Looking for related papers…
Contents