Paper deep dive
PixelSmile: Toward Fine-Grained Facial Expression Editing
Jiabin Hua, Hengyuan Xu, Aojie Li, Wei Cheng, Gang Yu, Xingjun Ma, Yu-Gang Jiang
Intelligence
Status: succeeded | Model: google/gemini-3.1-flash-lite-preview | Prompt: intel-v1 | Confidence: 95%
Last extracted: 3/27/2026, 1:39:22 AM
Summary
PixelSmile is a diffusion-based framework for fine-grained facial expression editing that addresses semantic overlap and identity preservation. It introduces the Flex Facial Expression (FFE) dataset with continuous affective annotations and the FFE-Bench for evaluation. The framework utilizes fully symmetric joint training and textual latent interpolation to achieve precise, linear, and continuous expression control.
Entities (5)
Relation Signals (3)
PixelSmile → implements → Fully Symmetric Joint Training
confidence 100% · PixelSmile, a diffusion framework that disentangles expression semantics via fully symmetric joint training.
PixelSmile → utilizes → FFE
confidence 95% · Building upon this data-centric foundation, we propose PixelSmile
FFE-Bench → evaluates → PixelSmile
confidence 90% · establish FFE-Bench to evaluate... editing accuracy
Cypher Suggestions (0)
No Cypher suggestions yet.
Abstract
Abstract:Fine-grained facial expression editing has long been limited by intrinsic semantic overlap. To address this, we construct the Flex Facial Expression (FFE) dataset with continuous affective annotations and establish FFE-Bench to evaluate structural confusion, editing accuracy, linear controllability, and the trade-off between expression editing and identity preservation. We propose PixelSmile, a diffusion framework that disentangles expression semantics via fully symmetric joint training. PixelSmile combines intensity supervision with contrastive learning to produce stronger and more distinguishable expressions, achieving precise and stable linear expression control through textual latent interpolation. Extensive experiments demonstrate that PixelSmile achieves superior disentanglement and robust identity preservation, confirming its effectiveness for continuous, controllable, and fine-grained expression editing, while naturally supporting smooth expression blending.
Tags
Links
- Source: https://arxiv.org/abs/2603.25728v1
- Canonical: https://arxiv.org/abs/2603.25728v1
Trouble viewing inline? Open PDF directly →
Full Text
71,699 characters extracted from source content.
Expand or collapse full text
PixelSmile: Toward Fine-Grained Facial Expression Editing Jiabin Hua 1,2,∗ Hengyuan Xu 1,2,∗ Aojie Li 2,† Wei Cheng 2 Gang Yu 2,‡ Xingjun Ma 1,‡ Yu-Gang Jiang 1 1 Fudan University 2 StepFun Project PageCodeModelBenchmarkDemo Real Human Anime Happy Happy Sad Disgust Angry Fear Surprised Anxious Sleepy Contempt Shy Confused confident I. Fine-Grained Expression EditingII. Extended Expression I. Expression Blending Figure 1. Overview of PixelSmile. It enables 1) continuous and precise control of facial expression intensity across real-world and anime domains, 2) editing across 12 distinct expression categories, and 3) seamless blending of multiple expressions. Abstract Fine-grained facial expression editing has long been limited by intrinsic semantic overlap. To address this, we construct the Flex Facial Expression (FFE) dataset with continuous affective annotations and establish FFE-Bench to evalu- ate structural confusion, editing accuracy, linear control- lability, and the trade-off between expression editing and identity preservation. We propose PixelSmile, a diffusion ∗ Equal contribution. † Project lead. ‡ Corresponding authors. framework that disentangles expression semantics via fully symmetric joint training. PixelSmile combines intensity su- pervision with contrastive learning to produce stronger and more distinguishable expressions, achieving precise and stable linear expression control through textual latent in- terpolation. Extensive experiments demonstrate that Pix- elSmile achieves superior disentanglement and robust iden- tity preservation, confirming its effectiveness for continu- ous, controllable, and fine-grained expression editing, while naturally supporting smooth expression blending. arXiv:2603.25728v1 [cs.CV] 26 Mar 2026 1. Introduction Recent advances in diffusion-based image editing mod- els [46, 73] and identity-consistent generation tech- niques [27, 34, 74] have significantly improved the ability to manipulate personal portraits using natural language. De- spite this progress, fine-grained facial expression editing re- mains a challenging problem. Current models can generate clearly distinct expressions, such as happy versus sad, but struggle to delineate highly correlated, semantically over- lapping expression pairs, such as fear versus surprise or anger versus disgust. Most existing methods rely on dis- crete expression categories, forcing inherently continuous human expressions into rigid class boundaries. As a result, these formulations fail to capture subtle expression bound- aries, leading to structured cross-category confusion, lim- ited control over expression intensity, and degraded identity consistency during editing. To better understand this limitation, we analyze the se- mantic structure of facial expressions. As illustrated in Fig. 2, facial expressions lie on a continuous semantic man- ifold where semantically adjacent emotions naturally over- lap. This overlap manifests as systematic confusion across multiple stakeholders: human annotators, classifiers, and generative models often fail to uniquely distinguish se- mantically adjacent expressions like fear versus surprise or anger versus disgust. When generative models are trained using discrete and potentially conflicting labels from such ambiguous samples, they are forced to learn entangled rep- resentations in the latent space. Consequently, this struc- tural entanglement prevents precise control, resulting in un- intended expression leakage, where editing one emotion in- advertently triggers the characteristics of another or even degrades identity consistency. Addressing this challenge requires a new supervision paradigm for facial expression editing models. Conven- tional datasets often represent facial expressions using rigid one-hot labels, which fail to capture the nuanced structure of human affect and propagate semantic entanglement into the generative pipeline. To address this limitation, we in- troduce a new supervision paradigm based on continuous affective annotations. Specifically, we construct the Flex Facial Expression (FFE) dataset, which replaces discrete labels with continuous 12-dimensional affective score dis- tributions. Based on this dataset, we further establish FFE- Bench to evaluate structural confusion, editing accuracy, linear controllability, and the trade-off between expression editing and identity preservation. By providing diverse ex- pressions within the same identity and continuous affec- tive ground truth across both real and anime domains, FFE breaks the one-hot supervision bottleneck, allowing models to learn the fine-grained boundaries of the expression mani- fold rather than disjoint categories, and enabling systematic evaluation of controllable expression editing. Building upon this data-centric foundation, we propose PixelSmile, a diffusion-based editing framework that dis- entangles expression semantics. Our framework introduces a fully symmetric joint training paradigm to contrast con- fusing expression pairs identified in our analysis. Com- bined with a flow-matching-based textual latent interpola- tion mechanism, PixelSmile enables precise and linearly controllable expression intensity at inference time without requiring reference images. Through the synergy between continuous affective supervision and symmetric learning, PixelSmile achieves robust and controllable editing while preserving identity fidelity. In summary, our contributions are threefold: • Systematic Analysis of Semantic Overlap. We reveal and formalize the structured semantic overlap between fa- cial expressions, demonstrating that structured semantic overlap, rather than purely classification error, is a pri- mary cause of failures in both recognition and generative editing tasks. • Dataset and Benchmark.We construct the FFE dataset—a large-scale, cross-domain collection featuring 12 expression categories with continuous affective anno- tations—and establish FFE-Bench, a multi-dimensional evaluation environment specifically designed to evaluate structural confusion, expression editing accuracy, linear controllability, and the trade-off between expression edit- ing and identity preservation. • PixelSmile Framework. We propose a novel diffusion- based framework utilizing fully symmetric joint training and textual latent interpolation. This design effectively disentangles overlapping emotions and enables disentan- gled and linearly controllable expression editing. 2. Related Work Facial Expression Editing.Facial expression editing aims to modify facial expressions while preserving iden- tity. Early approaches relied on conditional GANs [24], formulating the task as multi-domain image-to-image trans- lation [10, 11, 15, 45, 59]. Subsequent works explored disentangled latent manipulation within StyleGAN-based architectures [28, 36, 37, 64, 65, 80] to identify seman- tic directions for continuous expression control. Another line of research incorporates explicit facial priors, such as Action Units or 3DMM parameters, to enable structured, interpretable manipulation. For instance, MagicFace [71] leverages such priors to guide diffusion models, while other works [13, 16, 22, 33, 59] explore similar structural constraints. Despite facilitating discrete expression trans- fers, these methods often struggle with fine-grained control, identity consistency, and generalization. More recently, diffusion models [30] have significantly advanced image generation and editing quality [4, 29, 49, 82]. Further- more, large-scale multimodal pretraining has fueled signifi- SurprisedFearAngryDisgust Shared: Wide Eyes, Open Mouth Shared: Frown, Negative Affect Fear: 0.95 Surprised! Human Model Recognition Confusion Expression Semantic Overlap Editing Confusion AngryDisgust Angry Disgust FearSurprised FFE Dataset “id”: ‘001136’, “category”: ‘angry’, “scores”: “happy”: 0.85 , “angry”: 0.0 , “sad”: 0.1 , “surprised”: 0.02 , “fear”: 0.0 , “disgust”: 0.0 , “anxious”: 0.01 , “confident”: 0.55 , “contempt”: 0.01 , “confused”: 0.0 , “shy”: 0.02 , “sleepy”: 0.0 , , “box”:[53.21,...] Data Structure Angry Real Human Samples Anime Samples Key Observation Our Contribution PixelSmile Figure 2. Observation of Expression Semantic Overlap. Inherent expression overlap causes systematic confusion across human annota- tors, recognition models, and generative models (top). We resolve this via the FFE dataset (bottom left) and PixelSmile framework (bottom right), utilizing continuous supervision and symmetric training for disentangled editing. cant advancements in general-purpose editing. Large-scale foundation models, such as GPT-Image [54], Nano Banana Pro [25], Qwen-Image [73], and LongCat-Image [68], now demonstrate remarkable zero-shot flexibility and editing ca- pabilities [5, 39, 46]. Continuously Controlled Generation.Prior works achieve continuous editing by leveraging interpolatable subspaces within generative models. ConceptSlider [20] interpolates LoRA weights, while subsequent methods [3, 7, 12, 21, 23, 26, 29, 32, 35, 63, 67, 77, 85] manipulate text embeddings or modulation features to achieve grad- ual semantic variation.More recently, SliderEdit [81], Kontinuous-Kontext [57], and concurrent works [72, 75, 79] extend continuous control to editing models built upon FLUX.1 Kontext [40]. Despite smoother transitions via re- duced strength or pixel interpolation, these methods remain constrained by entangled latent spaces, leading to semantic ambiguity and identity drift at large magnitudes. By dis- entangling latent expression semantics, our structured for- mulation achieves fine-grained linear control and identity preservation across diverse manipulation strengths. Facial Expression Datasets and Benchmarks.High- quality datasets and reliable benchmarks are essential for facial expression analysis. Early controlled datasets [41, 47, 48, 78] provide same-identity multi-expression samples for precise comparison but lack diversity, while large-scale in- the-wild datasets [2, 42, 50, 76, 83] enhance generalization but lack paired expressions for the same identity, hinder- ing identity-expression disentanglement in generative edit- ing. Recent efforts extend to video and multimodal settings. While video-based datasets [51, 60, 84] focus on temporal or cross-modal dynamics, the MEAD dataset [69] provides expressions with three distinct intensity levels, moving be- yond purely categorical labels but still falling short of fine- grained, continuous control and structured disentanglement in static editing contexts. Alongside these, benchmarks such as F-Bench [44] and SEED [87] evaluate facial gen- eration using visual metrics and human preference. How- ever, standard metrics (e.g., CLIP, SSIM, LPIPS) capture overall quality but offer limited insight into disentanglement and continuous control. To address these gaps, we propose FFE and FFE-Bench. By providing same-identity pairs with continuous affective annotations, our approach enables rig- orous evaluation of fine-grained, linearly controllable, and disentangled expression editing. 3. Dataset and Benchmark To facilitate fine-grained and linearly controllable facial ex- pression editing, we construct the FFE dataset and estab- lish FFE-Bench, a dedicated evaluation benchmark. Ex- isting datasets often lack same-identity expression diver- sity or provide only discrete expression labels, which limits the evaluation of controllable expression manipulation. Our dataset addresses these limitations by providing large-scale same-identity expression variations with continuous affec- tive annotations, enabling systematic analysis of expression disentanglement and editing controllability. Flow Matching Loss 1 2 푻푮 fear ,푷 fear ,푵 surp +푻푮 surp ,푷 surp ,푵 fear GT(Surprise) GT(Fear) Gen(Fear) GT(Fear) GT(Surprise) Gen(Surprise) SourceImage Generated Surprise Generated Fear 1. He’s Surprised(0.8) 2. He is in Fear(0.6) GT Fear(0.6) GT Surprise(0.8) Source 1 − cos , AngrierNeutral InferenceStage Neutral Prompt Text Encoder Target Prompt Text Encoder PixelSmile Interpolated Embedding 푇 푖 =푇 푖푛푖 +훼 ⋅ (푇 target − 푇 푖푛푖 ) Contrastive Loss ID Loss Text Encoder DiT Blocks LoRA PixelSmile VAE Training Stage ArcFace Figure 3. Framework Overview. (1) Inference Stage. We interpolate between the neutral and target expression embeddings in textual latent space using a controllable coefficient α, enabling continuous adjustment of expression intensity. (2) Training Stage. We adopt a joint fully symmetric training framework. Specifically, we sample a source image P src and a confusing expression pair(P a ,P b ) to construct a triplet. We first treat P a as the positive and P b as the negative to compute a joint loss, and then swap their roles to compute it again, yielding a symmetric training objective. The joint loss consists of three components: a Flow-Matching loss for intensity alignment, a contrastive loss for expression separation, and an identity preservation loss to maintain subject consistency. 3.1. The FFE Dataset FFE is constructed through a four-stage collect–compose– generate–annotate pipeline designed to ensure expression diversity, cross-domain coverage, and reliable annotations. The final dataset contains 60,000 images across real and anime domains, supporting both photorealistic and stylized facial expression editing. Base Identity Collection. We first curate a set of high- quality base identities from two domains: (1) Real do- main: approximately 6,000 real-world portraits are col- lected from public portrait datasets [1, 66], covering di- verse demographics and scene compositions, including both close-up and full-body images; (2) Anime domain: to enable cross-domain evaluation, we collect stylized portraits from 207 anime productions covering 629 characters, from which around 6,000 high-quality images are retained after quality filtering and automated face detection. For both domains, automated face detection followed by manual verification is applied to ensure identity clarity and image quality. These images form the identity backbone of FFE dataset. Expression Prompt Composition. To obtain fine-grained expression variations, we construct a structured prompt li- brary for 12 target expressions. The taxonomy consists of six basic emotions [19] and six extended emotions (Con- fused, Contempt, Confident, Shy, Sleepy, Anxious). Rather than relying solely on abstract expression labels, each ex- pression is decomposed into facial attribute components (e.g., mouth shape, eyebrow movement, and eye openness). Candidate attribute combinations are automatically gener- ated and filtered with a vision-language model to remove anatomically inconsistent or semantically conflicting de- scriptions, resulting in a validated library of fine-grained expression prompts. Controlled Expression Generation. For each base iden- tity, multiple target expressions with varying intensities are synthesized using a state-of-the-art image editing model, Nano Banana Pro. We adopt a dual-part prompt design that specifies both the global expression category and localized facial attributes, improving controllability and reducing am- biguity between semantically similar expressions. This pro- cess produces approximately 60,000 images in total (30,000 per domain), providing rich identity-preserving expression variations across diverse conditions. Continuous Annotation and Quality Filtering. Departing from conventional one-hot expression labels, each image is annotated with a 12-dimensional continuous score vector v ∈ [0, 1] 12 . The scores are predicted by a vision-language model, Gemini 3 Pro, which estimates the intensity of each expression category. A subset of samples is verified by hu- man annotators to ensure reliability. This representation captures semantic overlap between facial expressions (e.g., fear and surprise), providing a faithful approximation of the affective manifold. We further perform consistency checks and manual spot verification to remove ambiguous or low- confidence samples. The resulting dataset provides same- identity expression variations with continuous soft labels, enabling fine-grained evaluation of expression disentangle- ment and controllable facial expression editing. 3.2. The FFE-Bench Benchmark Motivated by the intrinsic semantic entanglement among fa- cial expressions, which leads to structured cross-category confusion, we design a unified benchmark to evaluate fa- cial expression editing from four complementary aspects: structural confusion, the trade-off between expression edit- ing and identity preservation, control linearity, and expres- sion editing accuracy. All expression classifications and in- tensity scores are predicted by Gemini 3 Pro. Mean Structural Confusion Rate (mSCR). To quantify structured confusion between semantically similar expres- sions, we define the directed confusion rate C i→j and the bidirectional confusion rate (BCR) as follows: C i→j = 1 N i N i X k=1 1(ˆy (i) k = j),(1) BCR(i,j) = 1 2 (C i→j + C j→i ),(2) where N i denotes the number of samples edited toward class i, and ˆy (i) k is the predicted dominant expression. The mSCR is computed by averaging BCR(i,j) over predefined confusing pairs (e.g., Fear–Surprise and Angry–Disgust). A lower mSCR indicates reduced cross-category confusion and improved semantic disentanglement. Harmonic Editing Score (HES). Facial expression edit- ing requires both accurate expression transfer and identity preservation. We define the Harmonic Editing Score as HES = 2× S E × S ID S E + S ID ,(3) where S E denotes the VLM-based target expression score, and S ID is the cosine similarity between source and edited faces. Identity similarity is computed as the average cosine similarity from three face recognition models (including Ar- cFace [14], AdaFace [38], FaceNet [62]) for robustness. High HES is achieved only when both expression strength and identity fidelity are preserved. Control Linearity Score (CLS). To evaluate continuous controllability, we feed uniformly spaced intensity coeffi- cients α ∈ [0,α max ] during inference and compute the Pearson correlation between α and the VLM-predicted in- tensity scores. Higher CLS indicates more linear and pre- dictable expression control. Expression Editing Accuracy (Acc). We report the pro- portion of generated images whose predicted dominant ex- pression matches the target instruction. This metric mea- sures overall categorical editing success. 4. Method We present PixelSmile, a framework for fine-grained facial expression editing. As illustrated in Fig. 3, our method 0.10.20.30.40.50.60.7 Expression Score 0.4 0.5 0.6 0.7 0.8 0.9 ID Sim Narrower ID Impairment Wider Expression Range Ours (Full) KSlider SliderEdit Ours(w/o CL) 0.0 3.0 0.0 3.0 0.0 3.0 0.0 3.0 Figure 4. Quantitative Evaluation of Linear Control Methods. Comparison of the trade-off between ID similarity and expression score across different models. PixelSmile achieves an optimal bal- ance, providing a wider expression manipulation range while pre- serving identity fidelity. builds upon a pretrained Multi-Modal Diffusion Trans- former (MMDiT) [58] with LoRA adaptation [31]. To ad- dress intrinsic semantic entanglement and enable continu- ous intensity control, we introduce two key components: (1) a Flow-Matching-based textual interpolation mecha- nism [43] for smooth expression strength control; and (2) a Fully Symmetric Joint Training framework with a symmet- ric contrastive objective to reduce cross-category confusion while preserving identity and background consistency. 4.1. Textual Latent Interpolation for Continuous Editing Existing expression editing approaches typically rely on discrete labels or coarse reference signals [73], which lim- its fine-grained control over expression intensity. Instead, we perform linear interpolation in the textual latent space to enable continuous and smooth expression manipulation. Textual Latent Interpolation. Given a neutral prompt P neu and a target expression prompt P tgt , the frozen MMDiT text encoder maps them to embeddings e neu and e tgt , respectively. We define the residual direction ∆e = e tgt − e neu ,(4) which captures the semantic shift from neutral to the target expression. A continuous conditioning embedding is then con- structed as e cond (α) = e neu + α· ∆e, α∈ [0, 1].(5) When α = 0, the conditioning corresponds to neutral ex- pression; when α = 1, it recovers the full target expression. Intermediate values of α yield smoothly varying expression intensities. Importantly, the same direction also supports Table 1. Quantitative Evaluation of General Editing Models. Best, second best, and third best results are indicated by,, and respectively. MethodmSCR↓Acc-6↑Acc-12↑ID Sim↑ Closed Seedream [5]0.37250.52940.37370.7221 Nano Banana Pro [25]0.17540.84310.62000.7107 GPT-Image [54]0.11070.80390.63000.5056 Open FLUX-Klein [39]0.28500.45100.33100.4146 LongCat [68]0.17540.62750.41000.6036 Qwen-Edit [73]0.26250.45100.29000.6938 Ours(w/o training)0.24000.52940.35000.6769 Ours0.05500.86270.60000.6522 Table 2. Quantitative Evaluation of Linear Control Models. Best, second best, and third best results are indicated by,, and respectively. MethodCLS-6↑CLS-12↑ID Sim↑HES↑ SAEdit [35]-0.01830.0007-- ConceptSlider [20]0.3161 ∗ -0.62500.3656 AttributeControl [3] 0.2856 ∗ -0.36090.2712 K-Slider [57]-0.0459-0.06340.79740.3272 SliderEdit [81]0.55990.52170.74140.3441 Ours(w/o training) 0.68920.52170.67690.4086 Ours0.80780.73050.65220.4723 ∗ Evaluated on CLS-2 (happy, surprised). extrapolation: at inference time, α > 1 enables stronger ex- pression transfer while maintaining structural consistency. Score-Supervised Flow Matching. To enforce consistency between textual interpolation and visual intensity, we intro- duce score supervision during Flow Matching (FM) train- ing. Each training image is associated with a ground-truth intensity coefficient α gt ∈ [0, 1], derived from the continu- ous expression annotations. During LoRA fine-tuning, we set α = α gt and use e cond (α) as the conditioning input to the dual-stream attention blocks. The score-supervised ve- locity loss is defined as L edit FM =E t,x 0 ,x 1 h v θ (x t ,t,e cond (α))− (x 1 −x 0 ) 2 2 i , (6) where x 0 denotes the source image latent and x 1 denotes the edited target latent. This objective explicitly couples the in- terpolation coefficient with the corresponding visual trans- formation. At inference, continuous control is achieved by varying α, without requiring reference images. 4.2. Fully Symmetric Joint Training for Disentan- glement As stated in Sec. 1 and illustrated in Fig. 2, facial expres- sions lie on a continuous and highly overlapping seman- tic manifold. For example, Surprise and Fear share simi- lar arousal and facial cues, leading to structural confusion near class boundaries when trained with discrete supervi- sion only. Inspired by contrastive learning and the idea of symmetric learning [70], we introduce a Fully Symmetric Joint Training framework with a symmetric contrastive ob- jective in the feature space. Symmetric Construction. Given a pair of semantically overlapping expressions, (E a ,E b ), defined based on the confusion patterns observed in the FFE dataset, and an in- put image, the model performs two parallel generations, G a and G b , conditioned on prompts corresponding to E a and E b , respectively. For G a , the ground-truth image with ex- pression E a , denoted as P a , serves as the positive, while the image with expression E b , denoted as P b , is treated as a hard negative; the roles are reversed for G b . This symmetric design avoids directional bias and enforces consistent sepa- ration between confusing expressions. Symmetric Contrastive Loss. All images are encoded us- ing a frozen CLIP image encoder to capture expression se- mantics. The symmetric loss is defined as L SC = 1 2 [T (G a ,P a ,P b ) +T (G b ,P b ,P a )],(7) whereT pulls the generated sample toward its target while pushing it away from the confusing expression. We investigate three realizations of T , including hinge- based [62], log-ratio [52], and InfoNCE-style [53] formu- lations. In practice, we primarily adopt the InfoNCE-style objective due to its stable optimization. Detailed formula- tions and ablations are provided in the Appendix A. 4.3. Identity Preservation Strong intensity extrapolation (α > 1) or contrastive forces may degrade identity consistency. To stabilize biometric features, we introduce an identity preservation loss based on a pretrained face recognition model. Specifically, we adopt ArcFace [14] as a frozen identity encoder Φ arc (·). For generated images G a ,G b and their corresponding ground truths P a ,P b , the identity loss is defined as L ID = 1 2 X i∈a,b [1− cos(Φ arc (G i ), Φ arc (P i ))],(8) This term enforces identity consistency while allowing expression variation. 4.4. Overall Training Objective We fine-tune the LoRA parameters of the frozen MMDiT under a symmetric dual-branch training scheme, where a pair of confusing expressions (a,b) is optimized jointly for the same subject. The overall objective is defined as L total = 1 2 L a FM +L b FM + λ sc L SC + λ id L ID ,(9) where λ sc and λ id control the trade-off between disentan- glement and identity preservation. This symmetric formu- lation jointly enforces continuous intensity control, expres- sion separation, and identity consistency. Angry Disgust Angry Disgust Fear Surprised Fear Surprised Nano Banana Pro GPT-Image 1.5 Seedream- 4.5 FLUX.2 Klein LongCat- Image-Edit Qwen-Image- Edit-2511 Ours Figure 5. Qualitative Comparison with General Editing Models. PixelSmile produces clearer expression changes while preserving facial identity, whereas existing editing models either weaken expression editing or degrade identity consistency. 5. Experiment 5.1. Experimental Setup We implement PixelSmile based on Qwen-Image-Edit- 2511. To handle the distinct stylistic distributions of real- world and anime domains, we train two independent LoRA adapters for each. Following prior work [6, 88], for con- trastive supervision, we adopt CLIP-ViT-L/14 [61] for the real domain and DanbooruCLIP [55] for anime. Identity preservation is enforced using a pretrained ArcFace (an- telopev2) model for the real domain. Additional implemen- tation details are provided in Appendix B. Baselines. To ensure a comprehensive and fair evaluation, we divide baselines into two groups according to their pri- mary strengths in facial expression editing: general editing models, which are strong in overall expression editing qual- ity, and linear control models, which are designed for con- tinuous and predictable intensity control. Concept Slider Attribute Control SliderEdit K-Slider Ours Concept Slider Attribute Control SliderEdit K-Slider Ours Figure 6. Qualitative Comparison with Linear Control Models. PixelSmile achieves smooth and monotonic expression transitions while preserving facial identity, whereas existing control methods either produce unstable responses or sacrifice identity consistency. The figure illustrates two representative expressions: happy (top row) and surprised (bottom row). Group 1: General Editing Models. This group repre- sents the strongest general-purpose text-guided image edit- ing systems. We include three closed-source commercial systems: Nano Banana Pro, GPT-Image-1.5 (GPT-Image), Seedream-4.5 (Seedream), and three open-source mod- els: Qwen-Image-Edit-2511 (Qwen-Edit), FLUX.2 Klein (FLUX-Klein), and LongCat-Image-Edit (LongCat). In the following, we refer to each model by the abbreviated name in parentheses.Although these models do not provide explicit mechanisms for fine-grained linear control, their strong generative priors make them competitive in overall expression editing quality. We therefore use them to eval- uate expression editing accuracy and the ability to resolve structural confusion between semantically overlapping ex- pressions. Group 2: Linear Control Models. This group fo- cuses on continuous attribute manipulation in latent space. We compare with recent control-oriented editing models including Kontinuous-Kontext (K-Slider), SliderEdit, and SAEdit. We also include earlier latent control approaches ConceptSlider and AttributeControl, using their officially recommended inversion strategies for real-image editing. While these earlier methods pioneered latent attribute con- trol, they are often limited by narrow predefined attribute categories and information loss introduced by inversion. We therefore treat them as reference baselines rather than pri- mary competitors in multi-category quantitative evaluation. w/o ID LossOurssrc Figure 7. Ablation on identity loss. Without ID loss, large ex- pression intensities cause identity drift in hairstyle and skin tex- ture. Our full method preserves identity consistently. Evaluation Metrics. We adopt the benchmark protocol de- fined in Sec. 3.2. For Group 1, we evaluate editing accuracy and expression disentanglement using Acc-6, Acc-12, and mSCR. For Group 2, we evaluate linear intensity control and identity fidelity using CLS-6, CLS-12, and HES. 5.2. Quantitative Evaluation We quantitatively compare PixelSmile with both general editing and linear control models in Table 1 and Table 2. Evaluation with General Editing Models. As shown in Table 1, we evaluate baselines on editing accuracy, struc- tural confusion, and identity fidelity. For the six basic ex- pressions, PixelSmile achieves the highest editing accuracy (0.8627), surpassing Nano Banana Pro (0.8431) and GPT- Image (0.8039). On the twelve extended expressions, our method remains among the best-performing models. This partially reflects the bias of the VLM scoring model (Gem- ini 3 Pro), which is highly reliable on basic expressions but less consistent on extended expression categories. More im- portantly, PixelSmile achieves the lowest structural confu- sion rate (0.0550), significantly outperforming GPT-Image (0.1107) and Nano Banana Pro (0.1754), while most other models exceed 0.2000. A value approaching 0.5 indicates that the model tends to collapse the confusing expression pair into a single expression, reflecting poor disentangle- ment of overlapping expressions. In terms of identity fi- delity, empirical observations in [74] suggest that realistic facial expression editing typically yields ID similarity val- ues around 0.6–0.7. Scores above 0.8 often indicate rigid “copy-paste” behavior, while scores below 0.5 imply se- vere identity distortion. Some baselines fall into these ex- tremes: Seedream maintains high ID similarity but suffers from large structural confusion due to limited edits, whereas FLUX-Klein drops below 0.5, significantly degrading iden- tity consistency. In contrast, PixelSmile produces strong ex- pressions while maintaining identity similarity within the natural range, achieving a better balance between expres- sion strength and identity preservation. Evaluation with Linear Control Models.PixelSmile demonstrates robust and consistent linear controllability across all metrics. ConceptSlider and AttributeControl are limited to editing only two expression attributes (happy and surprised) and produce weak editing effects; therefore, we report them as reference baselines with partial met- rics (e.g., CLS-2) rather than full-category comparisons. SAEdit is a text-to-image method that does not explicitly support identity-preserving editing; therefore, we include it only for quantitative reference and do not provide detailed qualitative analysis. As shown in Table 2 and Figure 4, simply applying textual embedding interpolation to Qwen- Edit (zero-shot) already yields competitive controllability (CLS-6 0.6892, HES 0.4086), outperforming existing con- trol baselines. With the proposed symmetric joint training, w/o Contrastive Lossw/o Symmetrical frameworkOurs Figure 8. Ablation on symmetric contrastive learning. Both w/o Contrastive Loss and w/o Symmetric Framework suffer from expression confusion, while our full method achieves precise expression disentanglement. The upper three rows show angry and disgust, and the lower three rows show fear and surprised. PixelSmile further improves performance and achieves the best results across all benchmarks (CLS-6 0.8078, CLS-12 0.7305, and HES 0.4723), indicating that explicitly mod- eling expression semantics is critical for stable and fine- grained controllability. Figure 4 further reveals the limi- tations of existing methods. Although K-Slider and Slid- erEdit maintain high average ID similarity, this is largely because low editing intensities produce negligible changes, yielding ID similarity values close to 1.0. Specifically, K- Slider exhibits negative CLS scores and irregular intensity fluctuations that never exceed∼0.3, failing to establish lin- ear controllability. SliderEdit shows increasing expression intensity but forces a rapid drop in ID similarity (down to ∼0.4) once expression scores approach 0.5, indicating a trade-off between editing strength and identity preserva- tion. In contrast, PixelSmile achieves a monotonic response across a wide intensity range (expression scores reaching ∼0.8) while maintaining identity similarity within the natu- ral 0.6–0.7 interval, effectively balancing controllability and fidelity. This behavior demonstrates that our method not only improves average performance but also ensures stable and predictable control across the intensity spectrum. Table 3. Ablation Study. Best, second best, and third best results are indicated by,, andrespectively. AblationmSCR↓ACC-6↑ACC-12↑CLS-6↑CLS-12↑HES↑ID Sim↑ Loss w/o Contrastive Loss0.27250.64710.58890.69780.58890.45000.7018 w/o ID Loss0.05500.88240.65000.82150.68740.44510.5749 w/o Sym. Frame.0.13500.78430.47000.79390.64880.42530.6402 Constraint w/ Log-Ratio Constraint0.17500.80390.53000.79170.65460.49330.6943 w/ Hinge Constraint0.09500.88240.66000.79970.72280.47580.6280 DataMEAD0.21250.76470.47000.70470.61870.42350.5735 OursFull Setting0.05500.86270.60000.80780.73050.47230.6522 5.3. Qualitative Comparison We qualitatively compare PixelSmile with both general editing models and linear control baselines, as illustrated in Figure 5 and Figure 6. Comparison with General Editing Models. As shown in Figure 5, existing general editing models struggle to simul- taneously achieve clear expression editing and strong iden- tity preservation. Several models, including Nano Banana Pro, Qwen-Edit, Seedream, and LongCat, preserve iden- tity well but produce only weak expression changes, of- ten resulting in barely noticeable edits. In contrast, GPT- Image generates more visible expression differences but introduces moderate identity drift. FLUX-Klein performs the worst in both aspects, showing weak expression editing while severely degrading identity consistency. Compared with these methods, PixelSmile produces clear and recog- nizable expression changes while maintaining stable facial identity, achieving the best balance between semantic edit- ing and identity preservation. Comparison with Linear Control Models. Figure 6 com- pares continuous expression control across different meth- ods. An ideal method should produce expression inten- sity that increases monotonically with the control param- eter while preserving identity consistency. We first analyze the relatively simple expression Happy. ConceptSlider and AttributeControl show limited linear response but quickly degrade identity as editing strength increases. SliderEdit exhibits a step-like behavior: expressions remain nearly un- changed for most control values and suddenly increase at higher strengths, accompanied by significant identity degra- dation. K-Slider shows unstable behavior, where expres- sion changes have little correlation with the control param- eter. When moving to the more challenging expression Sur- prised, the linear response of these methods further dete- riorates and identity preservation becomes worse. In con- trast, PixelSmile maintains a stable monotonic increase in expression intensity while preserving identity across the en- tire control range. Even for more difficult expressions such as Disgust, our method continues to produce clear and con- trollable expression changes. 5.4. Ablation Study To validate the necessity of each component in PixelSmile, we conduct comprehensive ablation experiments, with quantitative results summarized in Table 3. Overall, the re- sults reveal an inherent trade-off between expression editing capability and identity preservation: stronger editing often leads to identity degradation, while excessive identity con- straints suppress effective expression transfer. Ablation on Loss Framework. We first analyze the roles of the identity loss and the contrastive loss. Removing the identity loss improves expression editing and disentangle- ment but significantly degrades identity consistency. The model tends to modify facial attributes such as hairstyle or skin texture to match the target expression, especially at large editing intensities, leading to clear identity drift and inconsistent facial appearance across edits. As illustrated in Fig. 7, the full model maintains stable identity while the variant without ID loss shows noticeable identity changes, confirming the importance of identity supervision for pre- serving subject consistency. 10003000500070009000 Steps 0.06 0.08 0.10 0.12 0.14 mSCR w/o Sym. frame. (mSCR) Ours (mSCR) w/o Sym. frame. (Loss) Ours (Loss) 0.35 0.40 0.45 0.50 0.55 0.60 Train Loss Figure 9. Training dynamics of symmetric contrastive learn- ing. The asymmetric variant reduces loss faster in early train- ing but leads to higher structural confusion, while the symmetric framework achieves lower and more stable mSCR. Conversely, removing the contrastive loss yields the highest identity similarity but leads to the weakest edit- ing accuracy and the highest structural confusion. Without the contrastive objective, the model collapses toward recon- structing the source image instead of performing meaning- ful expression edits. As shown in Fig. 8, the model without contrastive supervision fails to separate semantically sim- ilar expressions, resulting in severe expression confusion. These results demonstrate that the two losses play comple- mentary roles: identity loss stabilizes facial identity, while contrastive loss enhances expression disentanglement. Ablation on Symmetric Framework. We further com- pare the proposed symmetric training design with an asym- metric variant that applies contrastive supervision to only one branch. As shown in Fig. 8, removing the symmet- ric structure again leads to noticeable expression confusion. From the training dynamics in Fig. 9, the asymmetric model shows faster initial loss reduction but converges to worse solutions with lower editing accuracy and higher confusion rates. In contrast, the symmetric design acts as a structural regularizer: although it slows early convergence, the bidi- rectional constraints stabilize optimization and lead to bet- ter disentangled representations. Ablation on Triplet Formulations. We also compare three triplet formulations: Log-ratio, Hinge, and InfoNCE. Log- ratio favors identity preservation but weakens expression editing, while Hinge maximizes editing strength at the cost of identity consistency. InfoNCE achieves the best balance between expression disentanglement and identity fidelity, and is therefore adopted as the default formulation. Ablation on Dataset. Finally, we evaluate the impact of training data by training the same architecture on the widely used MEAD dataset [69], with preprocessing details pro- vided in Appendix C. The MEAD-trained model consis- tently underperforms our full model across all metrics. This gap is mainly due to MEAD’s limited identity diversity and discrete intensity annotations, which restrict fine-grained expression modeling and semantic disentanglement. In con- trast, our FFE provides richer identity variation and continu- ous soft-label supervision, enabling more precise and robust expression editing in the wild. 5.5. User Study We conducted a user study with 2,400 images and 10 trained annotators who ranked three continuous editing methods on expression continuity and identity consistency. Mean scores (continuity, identity) are: PixelSmile (4.48, 3.80); K-Slider (1.36, 4.06); SliderEdit (3.16, 1.14). As illustrated in Fig- ure 10, human judgments are consistent with the machine- based evaluation. Overall, PixelSmile achieves the best balance, attaining the highest continuity while maintaining strong identity preservation. 5.6. Expression Blend Human facial behavior often involves compound expres- sions [17, 18, 56]. To explore whether such compositional- ity emerges in the learned representation, we perform pair- wise linear interpolation among six basic expressions, pro- ducing 15 zero-shot combinations. As shown in Fig. 12 in the Appendix, several pairs generate perceptually coherent compound expressions, suggesting that the learned emo- tion manifold is continuous and compositional. However, some combinations collapse into a single dominant expres- sion (e.g., Fear+Surprise) or produce unstable results due to physiological conflicts (e.g., Angry+Happy). Overall, 9 out of 15 combinations form plausible compound expressions, indicating that the learned representation supports linear composition while respecting implicit facial constraints and capturing meaningful compositional structure. 12345 Continuity 1 2 3 4 5 Identity Consistency Ours K-Slider SliderEdit Figure 10. User study results. We show the trade-off between identity preservation and continuity of editing, annotated by hu- man annotators. The size of the points indicates the HES scores of human annotators. 6. Conclusion In this paper, we present PixelSmile, a framework for ad- dressing semantic entanglement in facial expression edit- ing. By shifting from discrete supervision to the continuous expression manifold defined by FFE and evaluated through FFE-Bench, our approach enables precise and linearly con- trollable editing via symmetric joint training. Extensive ex- periments demonstrate effectiveness of PixelSmile in four dimensions: structural confusion, expression accuracy, lin- ear controllability, and identity preservation. Overall, this work establishes a standardized framework for fine-grained facial expression editing and advances research toward con- tinuous and compositional facial affect manipulation. Ethics Statement All data in FFE is collected from publicly available sources and used in compliance with their respective licenses and terms of use. The real-world subset is derived from existing public datasets (e.g., Human Images Dataset [66] and Mat- ting Human Dataset [1], both distributed under the MIT Li- cense), while the anime subset consists of stylized fictional characters from publicly available media. We do not collect or use any private or login-restricted data. Facial expres- sion editing is a dual-use technology that may pose risks, such as misuse in identity-related scenarios. Our work fo- cuses on expression manipulation and is intended for non- commercial academic research. We do not aim to alter iden- tity or enable deceptive applications. To mitigate potential risks, no personal metadata is retained, and the dataset is curated to exclude offensive content. We encourage respon- sible use of the dataset and models in compliance with ap- plicable laws, regulations, and ethical guidelines. References [1] aisegmentcn.Matting human datasets. https : //github.com/aisegmentcn/matting_human_ datasets, 2020. Accessed: 2026-03-03. 4, 13 [2] Emad Barsoum, Cha Zhang, Cristian Canton Ferrer, and Zhengyou Zhang. Training deep networks for facial expres- sion recognition with crowd-sourced label distribution. In ICMI, 2016. 3 [3] Stefan Andreas Baumann, Felix Krause, Michael Neumayr, Nick Stracke, Melvin Sevi, Vincent Tao Hu, and Bj ̈ orn Om- mer.Continuous, subject-specific attribute control in t2i models by identifying semantic directions. In CVPR, 2025. 3, 6 [4] Tim Brooks, Aleksander Holynski, and Alexei A Efros. In- structpix2pix: Learning to follow image editing instructions. In CVPR, 2023. 2 [5] ByteDance.Seedream 4.5: Advanced ai image genera- tion model. https://seed.bytedance.com/en/ seedream4_5, 2025. Accessed: 2026-03. 3, 6 [6] Jingjing Chang, Yixiao Fang, Peng Xing, Shuhan Wu, Wei Cheng, Rui Wang, Xianfang Zeng, Gang Yu, and Hai-Bao Chen. Oneig-bench: Omni-dimensional nuanced evalua- tion for image generation. arXiv preprint arXiv:2506.07977, 2025. 7 [7] Ta Ying Cheng, Prafull Sharma, Mark Boss, and Varun Jam- pani. Marble: Material recomposition and blending in clip- space. In CVPR, 2025. 3 [8] Wei Cheng, Su Xu, Jingtan Piao, Chen Qian, Wayne Wu, Kwan-Yee Lin, and Hongsheng Li. Generalizable neural performer: Learning robust radiance fields for human novel view synthesis. arXiv preprint arXiv:2204.11798, 2022. 16 [9] Wei Cheng, Ruixiang Chen, Siming Fan, Wanqi Yin, Keyu Chen, Zhongang Cai, Jingbo Wang, Yang Gao, Zhengming Yu, Zhengyu Lin, et al. Dna-rendering: A diverse neural actor repository for high-fidelity human-centric rendering. In ICCV, 2023. 16 [10] Yunjey Choi, Minje Choi, Munyoung Kim, Jung-Woo Ha, Sunghun Kim, and Jaegul Choo. Stargan: Unified genera- tive adversarial networks for multi-domain image-to-image translation. In CVPR, 2018. 2 [11] Yunjey Choi, Youngjung Uh, Jaejun Yoo, and Jung-Woo Ha. Stargan v2: Diverse image synthesis for multiple domains. In CVPR, 2020. 2 [12] Yusuf Dalva, Kavana Venkatesh, and Pinar Yanardag. Fluxs- pace: Disentangled semantic editing in rectified flow trans- formers. arXiv preprint arXiv:2412.09611, 2024. 3 [13] Radek Dan ˇ e ˇ cek, Michael J Black, and Timo Bolkart. Emoca: Emotion driven monocular face capture and animation. In CVPR, 2022. 2 [14] Jiankang Deng, Jia Guo, Niannan Xue, and Stefanos Zafeiriou. Arcface: Additive angular margin loss for deep face recognition. In CVPR, 2019. 5, 6 [15] Hui Ding, Kumar Sricharan, and Rama Chellappa. Exprgan: Facial expression editing with controllable expression inten- sity. In AAAI, 2018. 2 [16] Zheng Ding, Xuaner Zhang, Zhihao Xia, Lars Jebe, Zhuowen Tu, and Xiuming Zhang. Diffusionrig: Learning personalized priors for facial appearance editing. In CVPR, 2023. 2 [17] Shichuan Du and Aleix M Martinez. Compound facial ex- pressions of emotion: from basic research to clinical appli- cations. Dialogues in clinical neuroscience, 2015. 12 [18] Shichuan Du, Yong Tao, and Aleix M Martinez. Compound facial expressions of emotion. PNAS, 2014. 12 [19] Paul Ekman. An argument for basic emotions. Cognition & emotion, 1992. 4 [20] Rohit Gandikota, Joanna Materzy ́ nska, Tingrui Zhou, Anto- nio Torralba, and David Bau. Concept sliders: Lora adaptors for precise control in diffusion models. In ECCV, 2024. 3, 6 [21] Rohit Gandikota, Zongze Wu, Richard Zhang, David Bau, Eli Shechtman, and Nick Kolkin. Sliderspace: Decomposing the visual capabilities of diffusion models. In ICCV, 2025. 3 [22] Nicola Garau, Niccolo Bisagno, Piotr Br ́ odka, and Nicola Conci. Deca: Deep viewpoint-equivariant human pose esti- mation using capsule autoencoders. In ICCV, 2021. 2 [23] Daniel Garibi, Shahar Yadin, Roni Paiss, Omer Tov, Shiran Zada, Ariel Ephrat, Tomer Michaeli, Inbar Mosseri, and Tali Dekel. Tokenverse: Versatile multi-concept personalization in token modulation space. TOG, 2025. 3 [24] Ian J Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. NeurIPS, 2014. 2 [25] Google.Nano banana pro:High-fidelity ai im- age generation and editing model. https : / / w.androidcentral.com/apps-software/ai/ googles-nano-banana-pro-and-more, 2025. Ac- cessed: 2026-03. 3, 6 [26] Julia Guerrero-Viu, Milos Hasan, Arthur Roullier, Midhun Harikumar, Yiwei Hu, Paul Guerrero, Diego Gutierrez, Be- len Masia, and Valentin Deschaintre. Texsliders: Diffusion- based texture editing in clip space. In SIGGRAPH, 2024. 3 [27] Zinan Guo, Yanze Wu, Chen Zhuowei, Peng Zhang, Qian He, et al. Pulid: Pure and lightning id customization via contrastive alignment. NeurIPS, 2024. 2 [28] Erik H ̈ ark ̈ onen, Aaron Hertzmann, Jaakko Lehtinen, and Sylvain Paris. Ganspace: Discovering interpretable gan con- trols. NeurIPS, 2020. 2 [29] Amir Hertz, Ron Mokady, Jay Tenenbaum, Kfir Aberman, Yael Pritch, and Daniel Cohen-Or. Prompt-to-prompt im- age editing with cross attention control.arXiv preprint arXiv:2208.01626, 2022. 2, 3 [30] Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffu- sion probabilistic models. NeurIPS, 2020. 2 [31] Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen- Zhu, Yuanzhi Li, Shean Wang, Liang Wang, Weizhu Chen, et al. Lora: Low-rank adaptation of large language models. ICLR, 2022. 5 [32] Rahul Jain, Amit Goel, Koichiro Niinuma, and Aakar Gupta. Adaptivesliders: User-aligned semantic slider-based editing of text-to-image model output. In CHI, 2025. 3 [33] Wooseok Jang, Youngjun Hong, Geonho Cha, and Seungry- ong Kim. Controlface: Harnessing facial parametric control for face rigging. In CVPR, 2025. 2 [34] Liming Jiang, Qing Yan, Yumin Jia, Zichuan Liu, Hao Kang, and Xin Lu. Infiniteyou: Flexible photo recrafting while pre- serving your identity. In ICCV, 2025. 2 [35] Ronen Kamenetsky, Sara Dorfman, Daniel Garibi, Roni Paiss, Or Patashnik, and Daniel Cohen-Or. Saedit: Token- level control for continuous image editing via sparse autoen- coder. arXiv preprint arXiv:2510.05081, 2025. 3, 6 [36] Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative adversarial networks. In CVPR, 2019. 2 [37] Tero Karras, Samuli Laine, Miika Aittala, Janne Hellsten, Jaakko Lehtinen, and Timo Aila. Analyzing and improving the image quality of stylegan. In CVPR, 2020. 2 [38] Minchul Kim, Anil K Jain, and Xiaoming Liu. Adaface: Quality adaptive margin for face recognition.In CVPR, 2022. 5 [39] Black Forest Labs. FLUX.2: Frontier Visual Intelligence. https://bfl.ai/blog/flux-2, 2025. 3, 6 [40] Black Forest Labs, Stephen Batifol, Andreas Blattmann, Frederic Boesel, Saksham Consul, Cyril Diagne, Tim Dock- horn, Jack English, Zion English, Patrick Esser, et al. Flux. 1 kontext: Flow matching for in-context image generation and editing in latent space. arXiv preprint arXiv:2506.15742, 2025. 3 [41] Oliver Langner, Ron Dotsch, Gijsbert Bijlstra, Daniel HJ Wigboldus, Skyler T Hawk, and AD Van Knippenberg. Pre- sentation and validation of the radboud faces database. Cog- nition and Emotion, 2010. 3 [42] Shan Li, Weihong Deng, and JunPing Du. Reliable crowd- sourcing and deep locality-preserving learning for expres- sion recognition in the wild. In CVPR, 2017. 3 [43] Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximil- ian Nickel, and Matt Le. Flow matching for generative mod- eling. arXiv preprint arXiv:2210.02747, 2022. 5 [44] Lu Liu, Huiyu Duan, Qiang Hu, Liu Yang, Chunlei Cai, Tianxiao Ye, Huayu Liu, Xiaoyun Zhang, and Guangtao Zhai. F-bench: Rethinking human preference evaluation metrics for benchmarking face generation, customization, and restoration. In ICCV, 2025. 3 [45] Ming Liu, Yukang Ding, Min Xia, Xiao Liu, Errui Ding, Wangmeng Zuo, and Shilei Wen. Stgan: A unified selec- tive transfer network for arbitrary image attribute editing. In CVPR, 2019. 2 [46] Shiyu Liu, Yucheng Han, Peng Xing, Fukun Yin, Rui Wang, Wei Cheng, Jiaqi Liao, Yingming Wang, Honghao Fu, Chun- rui Han, et al. Step1x-edit: A practical framework for general image editing. arXiv preprint arXiv:2504.17761, 2025. 2, 3 [47] Patrick Lucey, Jeffrey F Cohn, Takeo Kanade, Jason Saragih, Zara Ambadar, and Iain Matthews.The extended cohn- kanade dataset (ck+): A complete dataset for action unit and emotion-specified expression. In 2010 ieee computer soci- ety conference on computer vision and pattern recognition- workshops, 2010. 3 [48] Daniel Lundqvist, Anders Flykt, and Arne ̈ Ohman. Karolin- ska directed emotional faces. Cognition and Emotion, 1998. 3 [49] Chenlin Meng, Yutong He, Yang Song, Jiaming Song, Jia- jun Wu, Jun-Yan Zhu, and Stefano Ermon. Sdedit: Guided image synthesis and editing with stochastic differential equa- tions. arXiv preprint arXiv:2108.01073, 2021. 2 [50] Ali Mollahosseini, Behzad Hasani, and Mohammad H Ma- hoor. Affectnet: A database for facial expression, valence, and arousal computing in the wild. IEEE transactions on affective computing, 2017. 3 [51] Arsha Nagrani, Joon Son Chung, Weidi Xie, and Andrew Zisserman. Voxceleb: Large-scale speaker verification in the wild. Computer Speech & Language, 2020. 3 [52] Hyun Oh Song, Yu Xiang, Stefanie Jegelka, and Silvio Savarese. Deep metric learning via lifted structured feature embedding. In CVPR, 2016. 6 [53] Aaron van den Oord, Yazhe Li, and Oriol Vinyals. Repre- sentation learning with contrastive predictive coding. arXiv preprint arXiv:1807.03748, 2018. 6 [54] OpenAI. Introducing gpt-image-1.5. https://openai. com/index/new-chatgpt-images-is-here/, 2025. Accessed: 2026-03. 3, 6 [55] OysterQAQ. Danbooruclip. https://huggingface. co/OysterQAQ/DanbooruCLIP, 2023.Accessed: 2023-05-18. 7 [56] Dongwei Pan, Long Zhuo, Jingtan Piao, Huiwen Luo, Wei Cheng, Yuxin Wang, Siming Fan, Shengqi Liu, Lei Yang, Bo Dai, et al. Renderme-360: A large digital asset library and benchmarks towards high-fidelity head avatars. NeurIPS, 2023. 12, 16 [57] Rishubh Parihar, Or Patashnik, Daniil Ostashev, R Venkatesh Babu, Daniel Cohen-Or, and Kuan-Chieh Wang. Kontinuous kontext: Continuous strength control for instruction-based image editing. arXiv preprint arXiv:2510.08532, 2025. 3, 6 [58] William Peebles and Saining Xie. Scalable diffusion models with transformers. In ICCV, 2023. 5 [59] Albert Pumarola, Antonio Agudo, Aleix M Martinez, Al- berto Sanfeliu, and Francesc Moreno-Noguer. Ganimation: Anatomically-aware facial animation from a single image. In ECCV, 2018. 2 [60] Zongyang Qiu, Bingyuan Wang, Xingbei Chen, Yingqing He, and Zeyu Wang. Emovid: A multimodal emotion video dataset for emotion-centric video understanding and genera- tion. arXiv preprint arXiv:2511.11002, 2025. 3 [61] Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learn- ing transferable visual models from natural language super- vision. In ICLR, 2021. 7 [62] Florian Schroff, Dmitry Kalenichenko, and James Philbin. Facenet: A unified embedding for face recognition and clus- tering. In CVPR, 2015. 5, 6 [63] Prafull Sharma, Varun Jampani, Yuanzhen Li, Xuhui Jia, Dmitry Lagun, Fredo Durand, Bill Freeman, and Mark Matthews. Alchemist: Parametric control of material prop- erties with diffusion models. In CVPR, 2024. 3 [64] Yujun Shen and Bolei Zhou. Closed-form factorization of latent semantics in gans. In CVPR, 2021. 2 [65] Yujun Shen, Jinjin Gu, Xiaoou Tang, and Bolei Zhou. Inter- preting the latent space of gans for semantic face editing. In CVPR, 2020. 2 [66] snmahsa.Human images dataset (men and women). https://w.kaggle.com/datasets/snmahsa/ human-images-dataset-men-and-women, 2021. Accessed: 2026-03-03. 4, 13 [67] Deepak Sridhar and Nuno Vasconcelos. Prompt sliders for fine-grained control, editing and erasing of concepts in dif- fusion models. In ECCV, 2024. 3 [68] Meituan LongCat Team, Hanghang Ma, Haoxian Tan, Jiale Huang, Junqiang Wu, Jun-Yan He, Lishuai Gao, Songlin Xiao, Xiaoming Wei, Xiaoqi Ma, Xunliang Cai, Yayong Guan, and Jie Hu. Longcat-image technical report. arXiv preprint arXiv:2512.07584, 2025. 3, 6 [69] Kaisiyuan Wang, Qianyi Wu, Linsen Song, Zhuoqian Yang, Wayne Wu, Chen Qian, Ran He, Yu Qiao, and Chen Change Loy. Mead: A large-scale audio-visual dataset for emotional talking-face generation. In ECCV, 2020. 3, 12, 16 [70] Yisen Wang, Xingjun Ma, Zaiyi Chen, Yuan Luo, Jinfeng Yi, and James Bailey. Symmetric cross entropy for robust learning with noisy labels. In ICCV, 2019. 6 [71] Mengting Wei, Tuomas Varanka, Xingxun Jiang, Huai-Qian Khor, and Guoying Zhao. Magicface: High-fidelity facial expression editing with action-unit control. arXiv preprint arXiv:2501.02260, 2025. 2 [72] Alon Wolf, Chen Katzir, Kfir Aberman, and Or Patashnik. Continuous control of editing models via adaptive-origin guidance. arXiv preprint arXiv:2602.03826, 2026. 3 [73] Chenfei Wu, Jiahao Li, Jingren Zhou, Junyang Lin, Kaiyuan Gao, Kun Yan, Sheng-ming Yin, Shuai Bai, Xiao Xu, Yilei Chen, et al. Qwen-image technical report. arXiv preprint arXiv:2508.02324, 2025. 2, 3, 5, 6 [74] Hengyuan Xu, Wei Cheng, Peng Xing, Yixiao Fang, Shuhan Wu, Rui Wang, Xianfang Zeng, Daxin Jiang, Gang Yu, Xingjun Ma, et al.Withanyone: Towards control- lable and id consistent image generation. arXiv preprint arXiv:2510.14975, 2025. 2, 9 [75] Zhenyu Xu, Xiaoqi Shen, Haotian Nan, and Xinyu Zhang. Numerikontrol: Adding numeric control to diffusion trans- formers for instruction-based image editing. arXiv preprint arXiv:2511.23105, 2025. 3 [76] Yuqi Yang, Dongliang Chang, Yuanchen Fang, Yi-Zhe SonG, Zhanyu Ma, and Jun Guo. Controllable-continuous color editing in diffusion model via color mapping. arXiv preprint arXiv:2509.13756, 2025. 3 [77] Weixin Ye, Hongguang Zhu, Wei Wang, Yahui Liu, and Mengyu Wang. All-in-one slider for attribute manipulation in diffusion models. arXiv preprint arXiv:2508.19195, 2025. 3 [78] Lijun Yin, Xiaozhou Wei, Yi Sun, Jun Wang, and Matthew J Rosato. A 3d facial expression database for facial behavior research. In 7th international conference on automatic face and gesture recognition (FGR06), 2006. 3 [79] Xingxi Yin, Jingfeng Zhang, Yue Deng, Zhi Li, Yicheng Li, and Yin Zhang.Instructattribute: Fine-grained ob- ject attributes editing with instruction.arXiv preprint arXiv:2505.00751, 2025. 3 [80] O ̆ guz Kaan Y ̈ uksel, Enis Simsar, Ezgi G ̈ ulperi Er, and Pinar Yanardag. Latentclr: A contrastive learning approach for unsupervised discovery of interpretable directions. In ICCV, 2021. 2 [81] Arman Zarei, Samyadeep Basu, Mobina Pournemat, Sayan Nag, Ryan Rossi, and Soheil Feizi. Slideredit: Continuous image editing with fine-grained instruction control. arXiv preprint arXiv:2511.09715, 2025. 3, 6 [82] Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. In ICCV, 2023. 2 [83] Zhanpeng Zhang, Ping Luo, Chen Change Loy, and Xiaoou Tang. From facial expression recognition to interpersonal relation prediction. IJCV, 2018. 3 [84] Zhicheng Zhang, Weicheng Wang, Yongjie Zhu, Wenyu Qin, Pengfei Wan, Di Zhang, and Jufeng Yang. Videmo: Affective-tree reasoning for emotion-centric video founda- tion models. arXiv preprint arXiv:2511.02712, 2025. 3 [85] Weizhi Zhong, Huan Yang, Zheng Liu, Huiguo He, Zijian He, Xuesong Niu, Di Zhang, and Guanbin Li. Mod-adapter: Tuning-free and versatile multi-concept personalization via modulation adapter. arXiv preprint arXiv:2505.18612, 2025. 3 [86] Hao Zhu, Wayne Wu, Wentao Zhu, Liming Jiang, Siwei Tang, Li Zhang, Ziwei Liu, and Chen Change Loy. Celebv- hq: A large-scale video facial attributes dataset. In ECCV, 2022. 16 [87] Yule Zhu, Ping Liu, Zhedong Zheng, and Wei Liu. Seed: A benchmark dataset for sequential facial attribute editing with diffusion models. arXiv preprint arXiv:2506.00562, 2025. 3 [88] Cailin Zhuang, Ailin Huang, Yaoqi Hu, Jingwei Wu, Wei Cheng, Jiaqi Liao, Hongyuan Wang, Xinyao Liao, Wei- wei Cai, Hengyuan Xu, et al. Vistorybench: Comprehen- sive benchmark suite for story visualization. arXiv preprint arXiv:2505.24862, 2025. 7 Appendix A. Details of the Symmetric Contrastive Loss A.1. Triplet Constraint Formulations In this section, we provide detailed formulations of the triplet constraint function T (G,P,N ) used in Sec. 4. All features are extracted using a frozen CLIP image encoder to represent expression semantics and are ℓ 2 -normalized be- fore distance computation. For brevity, we denote d G,P = d(G,P ) and d G,N = d(G,N ) as cosine distances, and s G,P = sim(G,P ) and s G,N = sim(G,N ) as cosine simi- larities. Hinge-based Formulation. The margin-based objective is T hinge (G,P,N ) = max 0, d G,P − d G,N + m ,(10) where m is a fixed margin. Log-Ratio Formulation. We adopt a smooth distance- ratio objective: T ratio (G,P,N ) = log d G,P + ε d G,N + ε ,(11) where ε is a small constant for numerical stability. InfoNCE-style Formulation. The probabilistic con- trastive objective is T nce (G,P,N ) =− log exp(s G,P /τ ) P x∈P,N exp(s G,x /τ ) ,(12) where τ is a temperature parameter. A.2. Implementation Details Unless otherwise specified, we use the InfoNCE-style for- mulation with temperature τ = 0.07. For the hinge-based variant, the margin is set to m = 0.2. For the log-ratio formulation, we set ε = 10 −6 for numerical stability. All variants are evaluated under identical training schedules. B. Details of Experiment To ensure reproducibility and clarity, we provide additional implementation details for PixelSmile. Training is con- ducted on 4 NVIDIA H200 GPUs. LoRA Configuration. We apply LoRA to major attention and MLP components of the diffusion transformer. Key hy- perparameters are: rank = 64, α = 128, and dropout = 0. Training Hyperparameters. The models are optimized for 100 epochs using the AdamW optimizer with β 1 = 0.9, β 2 = 0.999, weight decay = 0.001, and ε = 1e−8. The learning rate is set to 1e−4 with cosine scheduling and 500 warmup steps. Mixed precision (bf16) is enabled to stabi- lize training. For the loss weights, we set λ SC = 1.0 (In- foNCE mode, symmetric) and λ ID = 0.1. The batch size per GPU is 4 with gradient accumulation steps = 1. C. Details of Dataset Ablation C.1. Dataset Overview Among human-centric dataset [8, 9, 56, 69, 86], we choose MEAD [69] to ablate on effectiveness of proposed dataset. The MEAD dataset [69] contains 7 discrete facial expres- sions captured from multi-view video sequences, with three intensity levels (low, medium, high). For our ablation, we only use the front-view subset and map its three intensity levels to continuous values 0.5,0.75,1.0 to match the input range of PixelSmile. C.2. Preprocessing and Triplet Construction Since MEAD provides video sequences, we uniformly sam- ple frames to obtain independent images. From these sam- pled frames, we construct triplets (P a ,P b ,I orig ) in the same manner as for FFE to train the symmetric contrastive framework. Each triplet consists of: • I orig 1: the source frame. • P a ,P b 2: two frames of the same subject with distinct expressions. Finally, we construct triplet data pairs from the same identities to conduct the symmetric contrastive training un- der our default configuration. D. Additional Qualitative Results This section provides additional qualitative results for Pix- elSmile. We present more examples of linear expression editing across multiple expression categories, as well as ad- ditional expression blending results obtained through inter- polation in the learned expression space. D.1. Additional Linear Expression Editing Results Figure 11 presents additional linear editing results for the remaining ten expressions across both real and anime do- mains. As the control parameter increases, the expression intensity changes smoothly while the facial identity remains consistent. D.2. Expression Blend Results Figure 12 shows the examples of expression blending ob- tained through pairwise interpolation between basic expres- sions. Anxious Contempt Disgust Fear Sad Angry Confident Confused Shy Sleepy Figure 11. Additional linear expression editing results. We show the remaining ten expressions across both real and anime domains. The top row shows results on real images, while the bottom row shows results on anime images. Expression intensity increases from left to right for each expression. HappySurprisedBlendedBlended BlendedBlended BlendedBlended HappySad HappyDisgust HappyFearFear DisgustSad Sad Figure 12. Expression Blending Results. Visualizing compositional facial expressions generated by smoothly blending multiple emotional categories in PixelSmile. E. Additional Dataset Details This section provides supplementary details of FFE, includ- ing annotation and scoring prompts, together with addi- tional dataset statistics. E.1. Annotation and Scoring Prompts We provide the prompt templates used in our annotation pipeline and expression scoring procedure. Table 4 and Table 5 present the prompts used for statistical annotation of the human and anime subsets, respectively. These two templates are designed to extract structured semantic at- tributes for dataset analysis and are both based on Qwen3- VL-235B-A22B. Table 6 shows the prompt used to assign expression intensity scores to images, which is based on Gemini 3 Pro. E.2. Dataset Statistics We present the statistical analysis of FFE in Fig. 13 and Fig. 14, covering categorical distributions and textual de- scription patterns across the real-world and anime domains. From Fig. 13, the real-world subset is diverse but im- balanced, dominated by young adults (53.5%), with chil- dren, teens, and seniors forming smaller proportions. Sim- ilar trends are observed in other attributes, where female samples are more frequent and light-to-medium skin tones constitute the majority, indicating that the dataset inherits non-uniform demographic characteristics. This bias reflects common patterns in portrait-centric internet images and in- troduces challenges for expression modeling. In contrast, the anime subset exhibits broader stylistic diversity, with CG and 2D anime each accounting for about 44%, along with additional styles such as chibi, manga, and sketch. Compared with the real-world subset, the anime subset also shows a flatter age distribution. However, it contains more unknown labels in attributes like gender and age, suggesting that stylized characters are inherently more ambiguous un- der real-world categorization schemes, which increases the difficulty of consistent annotation and evaluation. Fig. 14 further reveals domain-specific textual patterns. The real-world subset emphasizes natural appearance cues such as clothing, hairstyle, and facial details, while the anime subset contains more stylized and visually distinc- tive descriptions. This difference highlights that expres- sion editing in FFE involves both visual transformation and domain-dependent semantic interpretation, requiring mod- els to generalize across heterogeneous distributions. Over- all, these statistics indicate that FFE combines substan- tial diversity with realistic biases, making it a challenging and representative benchmark for fine-grained, diverse, and real-world facial expression editing. 0%10%20%30%40%50%60% Percentage (%) Young Adult Adult Middle Aged Child Teen Senior Unknown 53.5% 19.7% 12.4% 5.6% 5.0% 3.6% 0.1% (a) Age distribution in the real-world domain 0%10%20%30%40%50% Percentage (%) Cg Anime 2D Anime Chibi Manga Unknown Sketch Other 44.7% 44.1% 4.0% 2.7% 2.4% 1.2% 1.0% (b) Style distribution in the anime domain Figure 13. Statistical distributions of annotated data in FFE. The results provide insights into the underlying data characteristics across real-world and anime domains. Table 4. Human Dataset Annotation Prompt Template. Human Dataset Annotation Prompt System Context You are an image annotation assistant. For a single person image, output strict JSON only. Requirements • Describe only visible facts; do not infer identity, story, or intent. • categorical values must be selected from the provided enums. • All three fields in descriptions must be present. • Write all description sentences in English, each 8–25 words. JSON Schema "categorical": "gender": "male/female/androgynous/unknown", "age_group": "child/teen/young_adult/adult/middle_aged/senior/unknown", "skin_tone": "very_light/light/medium/dark/very_dark/unknown", "expression": "neutral/happy/sad/angry/surprised/fear/disgust/other/unknown" , "descriptions": "appearance_sentence": "One sentence about clothing and accessories.", "action_sentence": "One sentence about action or pose.", "background_sentence": "One sentence about scene/background." (a) Real-world appearance descriptions(b) Anime-style appearance descriptions Figure 14. Visualization of appearance-related textual descriptions in FFE. The visualizations highlight the distribution and diversity of annotations across real-world and anime domains. Table 5. Anime Dataset Annotation Prompt Template. Anime Dataset Annotation Prompt System Context You are an anime image annotation assistant. For a single character image, output strict JSON only. Requirements • Describe only visible facts; do not infer identity, story, or intent. • categorical values must be selected from the provided enums. • All three fields in descriptions must be present. • Write all description sentences in English, each 8–25 words. JSON Schema "categorical": "gender": "male/female/androgynous/unknown", "age_group": "child/teen/young_adult/adult/middle_aged/senior/unknown", "expression": "neutral/happy/sad/angry/surprised/fear/disgust/other/unknown", "anime_style": "2d_anime/chibi/manga/sketch/cg_anime/other/unknown" , "descriptions": "appearance_sentence": "One sentence about clothing and accessories.", "action_sentence": "One sentence about action or pose.", "background_sentence": "One sentence about scene/background." Table 6. Facial Expression Scoring Prompt Template. The same prompt is applied to both human and anime domains, highlighting the domain-agnostic nature of our scoring pipeline. Facial Expression Scoring Prompt Role and Task Definition You are an expert AI specialized in analyzing facial expressions in both photorealistic and anime-style images. Your task is to analyze the input image and determine the intensity score for specific expression categories. Target Emotion Definitions Analyze the image based on the following visual definitions. Crucially, pay close attention to the distinctions between similar expressions (e.g., Fear vs. Surprise, Anger vs. Disgust). • Happiness: Faintly smiling, cheerful, ecstatic; corners of the mouth raised or laughing. • Sadness: Somber, sorrowful, devastated; frowning with downcast eyes or tears. • Anger: Annoyed, angry, furious; furrowed brows and glaring eyes. • Fear: Wary, frightened, terrified; knitted brows and pupil constriction. • Surprise: Taken aback, amazed, stunned; eyes wide open or jaw dropped. • Disgust: Distasteful, repulsed, gagging; wrinkling the nose or raising the upper lip. • Embarrassment: A shy smile, blushing cheeks, avoiding eye contact, looking down, or covering the face with hands. • Confidence: Self-assured smile, corners of the mouth raised, sharp and firm gaze, slightly smug. • Confusion: Knitted brows, looking aside or rolling the eyes, head tilted in deep thought. • Drowsiness: Heavy drooping eyelids, yawning, lethargic or low-energy look. • Contempt: Asymmetrical smirk (one corner raised), sneering, looking down on someone. • Nervousness: Visible sweat drops, tense facial muscles, uneasy or restless gaze. Output Format Requirements • Return format: Output a standard JSON object. • Keys: Use the exact emotion labels provided above. • Values: The intensity score (a float between 0.00 and 1.00). • Constraint: Do not explain your reasoning. Output only the JSON object. Do not include Markdown formatting such as ‘json ... ‘. Output Example "Happiness": 0.90, "Surprise": 0.75, "Anger": 0.05, ...