Measured Sliders: Learning Continuous Controls from Differentiable Image Measurements

Yijia Chen; Boyu Wei; Xuanhua Yin

cs.CVSep 4, 2026

Measured Sliders: Learning Continuous Controls from Differentiable Image Measurements

Authors: Yijia Chen, Boyu Wei, Xuanhua Yin

Organizations: School of Computer Science, The University of Sydney

Abstract

Continuous sliders are useful only when coefficient changes produce predictable image changes. Yet most diffusion sliders derive their axes from text or learned representations, leaving their scales disconnected from observable image properties. Consequently, we cannot tell in advance which attributes are learnable, compare control strengths directly, or anticipate interference when multiple controls are combined. We propose Measured Sliders, a framework that defines continuous controls through closed-form differentiable image measurements. A common measurement space unifies the pipeline. Before training, an observability test identifies usable supervision. During training, a measurement-guided objective learns target movement while suppressing non-target changes. After training, decoded calibration expresses controls in comparable units of realized image change. Multiple LoRA branches are stored in one checkpoint and composed without training on joint activations. Across SDXL and FLUX.1-dev, the resulting controls are ordered, selective, and composable. On 553 prompts, lighting direction reaches rho = 0.995 and 98.9% monotone sweeps. A five-attribute checkpoint achieves average selectivity 2.59, compared with 1.50 for the strongest baseline, and preserves every requested direction in 96.7% of pair and 86.1% of triple compositions. The observability test also separates every subsequently successful measurement from the failed candidate. Overall, image-space measurement provides a common basis for learning, diagnosing, calibrating, and composing continuous generative controls.

Figures & tables

Figure 1: Measured Sliders turn image measurements into precise, disentangled controls. Left: in the matched FLUX setting, the text-defined slider is nearly flat, whereas measurement supervision follows the requested direction. Center: one checkpoint provides continuous brightness, saturation, warmth, contrast, and tint axes. Right: activating brightness produces the requested decoded change while the other four measurements remain nearly unchanged.

Figure 2: Measured Sliders pipeline. Observability: preview and decoded measurements are compared before training to retain candidates with a reliable optimization signal. Train: an active LoRA branch matches its target displacement while the remaining measurements are held still. Calibrate: decoded responses map raw coefficients to comparable image-space effect units for inference.

Method	Effect	Drift $\downarrow$	Sel. $\uparrow$	$\rho\uparrow$	Mono. $\uparrow$	DINO $\uparrow$
SDXL: decoded natural-variation units
Concept Sliders	0.039	0.085	0.453	0.122	33.5	0.924
Attr. Control	0.106	0.214	0.496	$-0.112$	18.4	0.810
Attr. Control, learned	0.101	0.254	0.395	0.117	24.6	0.818
FreeSliders	0.026	0.053	0.497	$-0.035$	26.2	0.944
Text Slider	0.020	0.037	0.537	$-0.077$	21.3	0.957

Table 1: Lighting-direction control on 553 GenEval prompts. All statistics use final decoded images. Effect and drift use fixed backbone-specific natural ranges and are not compared across panels. Monotonicity is reported as a percentage.

Figure 3: Pre-training observability predicts usable supervision. Each row pairs the observability ratio in the left panel with the subsequently measured decoded monotonicity in the right panel. Dashed lines mark the fixed retention threshold

\mathrm{SNR}_{k}=1

and learned-control threshold

\rho=0.8

. Five retained measurements exceed both thresholds. Sharpness falls below both.

Individually activated controls
Method	Bright. $K$	Satur. $S$	Warmth $W$	Contrast $C$	Tint $T$	Avg.
Concept Sliders	0.47	1.19	0.81	0.76	0.55	0.76
Attr. Control	2.18	1.10	1.08	0.73	0.76	1.17
FreeSliders †	1.41	1.47	1.61	0.86	2.14	1.50
Text Slider	1.81	0.99	1.36	0.64	1.06	1.17
Ours, one checkpoint	4.62	0.97	1.79	3.76	1.79	2.59

Table 2: Multi-control performance on six unseen contents. Upper: normalized selectivity of five branches in one checkpoint. Lower: zero-shot activation of all ten pairs and triples across every sign pattern and two seeds. All-sign success is a percentage.

Figure 4: Component analysis. Panels a through d evaluate natural-variation normalization, reachable targets, explicit stillness, and the frozen-text anchor. They test weak-coordinate balance, content consistency, non-target drift, and visual consistency, respectively.

Figure 5: Five continuous controls from one checkpoint. Calibrated branches share an unseen prompt, seed, noise, and sampler. Rows use $r_{k}\in\{-0.67,0,+0.67\}$ . Blue denotes the pixel-identical neutral.

Controllable Image Generation

Controllable image generation steers images using spatial, visual, semantic, stylistic, or physical constraints beyond ordinary text prompts.

See all 210 papers

Oct 5, 2026

UniSlider: Perceptually Uniform Sliders for Continuous Image Editing

Sliders provide an intuitive interface for continuous image editing. In current generative approaches, however, the slider is simply a rescaling of the method's strength parameter, such as an adapter coefficient, a prompt weight, or an interpolation factor. This strength relates poorly to perceptual change. The image can partially revert as the slider moves, long stretches of the range produce no visible difference, and short intervals transform the image abruptly. Remapping the strength could fix this uneven pace, but only if the trajectory is monotone, which current methods do not enforce. We therefore distinguish the slider from the strength, and require perceptual distance from the input to grow linearly with the slider value. We introduce UniSlider, a lightweight LoRA trained on a few-step editing backbone so that its strength approximates this ideal slider. Few-step sampling lets us impose this objective in pixel space without intermediate ground truth, and the backbone's output is preserved at full strength. However, a low-rank adapter cannot make the strength fully uniform. Our slider is thus an inference-time remapping of the strength, obtained by adaptive sampling. Since training optmizes to make the trajectory monotone, this remapping closes the remaining gap without extra training or parameters. On a new benchmark of 300 continuous edits evaluating uniformity, monotonicity, edit fidelity, and identity preservation, UniSlider outperforms all prior methods and is preferred in a user study.

David Serrano-Lozano, Duygu Ceylan, Yannick Hold-Geoffroy +3

Computer Vision Center · Universitat Autònoma de Barcelona · Adobe Research

Image Editing Evaluation Diffusion-Based Image Editing Image Editing

Ask PDF

Oct 9, 2026

SliderEdit++: Improved Slider Trajectories in Instruction-Based Image Editing

Slider-based control aims to give users continuous, fine-grained control over the intensity of individual edits in instruction-based image editing. However, existing approaches fail to deliver reliable continuous controllability: traversing a slider can produce abrupt visual jumps and non-monotonic changes, while jointly controlled sliders can interfere with one another. To systematically study these limitations, we introduce SlideBench, a comprehensive benchmark for continuous image editing spanning diverse editing categories and equipped with a unified suite of metrics quantifying monotonic progression, perceptual continuity, and inter-slider disentanglement. Building on these observations, we propose SliderEdit++, which explicitly regularizes the intermediate editing trajectory rather than supervising only its endpoints. Our method combines velocity- and semantic-space trajectory regularization with spectral control to encourage smooth, predictable slider traversal, together with a disentanglement objective that reduces interference between independently controlled edits. On SlideBench, SliderEdit++ achieves state-of-the-art performance, improving the overall score by 18.5% relative to the strongest baseline. At the same time, using a few-step editing backbone makes our method approximately 3x to 20x faster than existing slider-based approaches, enabling substantially more responsive interactive editing.

Arman Zarei, Sayan Nag, Bryan Wang +5

Adobe Research · University of Maryland

Image Editing Text-Guided Image Editing Image Editing Evaluation

Ask PDF

Apr 22, 2026

ParetoSlider: Diffusion Models Post-Training for Continuous Reward Control

Reinforcement Learning (RL) post-training has become the standard for aligning generative models with human preferences, yet most methods rely on a single scalar reward. When multiple criteria matter, the prevailing practice of ``early scalarization'' collapses rewards into a fixed weighted sum. This commits the model to a single trade-off point at training time, providing no inference-time control over inherently conflicting goals -- such as prompt adherence versus source fidelity in image editing. We introduce ParetoSlider, a multi-objective RL (MORL) framework that trains a single diffusion model to approximate the entire Pareto front. By training the model with continuously varying preference weights as a conditioning signal, we enable users to navigate optimal trade-offs at inference time without retraining or maintaining multiple checkpoints. We evaluate ParetoSlider across three state-of-the-art flow-matching backbones: SD3.5, FluxKontext, and LTX-2. Our single preference-conditioned model matches or exceeds the performance of baselines trained separately for fixed reward trade-offs, while uniquely providing fine-grained control over competing generative goals.

Shelly Golan, Michael Finkelson, Ariel Bereslavsky +2

Tel Aviv University · Lightricks · Adobe Research

Multi-Objective Reinforcement Learning RL for Image Generation RL for Diffusion Models RL Post-Training Conditional Generation

Ask PDF

Abstract

Controllable Image Generation

Explore similar work

UniSlider: Perceptually Uniform Sliders for Continuous Image Editing

SliderEdit++: Improved Slider Trajectories in Instruction-Based Image Editing

ParetoSlider: Diffusion Models Post-Training for Continuous Reward Control