Sparse Context achieves 2-4x faster inference in reference-conditioned diffusion models by fine-tuning with random token dropping and applying task-aware selection at inference time, without loss of visual quality.
hub Canonical reference
An Image is Worth One Word: Personalizing Text-to-Image Generation using Textual Inversion
Canonical reference. 93% of citing Pith papers cite this work as background.
98
Pith papers citing it
Background
93% of classified citations
abstract
Text-to-image models offer unprecedented freedom to guide creation through natural language. Yet, it is unclear how such freedom can be exercised to generate images of specific unique concepts, modify their appearance, or compose them in new roles and novel scenes. In other words, we ask: how can we use language-guided models to turn our cat into a painting, or imagine a new product based on our favorite toy? Here we present a simple approach that allows such creative freedom. Using only 3-5 images of a user-provided concept, like an object or a style, we learn to represent it through new "words" in the embedding space of a frozen text-to-image model. These "words" can be composed into natural language sentences, guiding personalized creation in an intuitive way. Notably, we find evidence that a single word embedding is sufficient for capturing unique and varied concepts. We compare our approach to a wide range of baselines, and demonstrate that it can more faithfully portray the concepts across a range of applications and tasks. Our code, data and new words will be available at: https://textual-inversion.github.io
hub tools
citation-role summary
background 15
citation-polarity summary
claims ledger
- abstract Text-to-image models offer unprecedented freedom to guide creation through natural language. Yet, it is unclear how such freedom can be exercised to generate images of specific unique concepts, modify their appearance, or compose them in new roles and novel scenes. In other words, we ask: how can we use language-guided models to turn our cat into a painting, or imagine a new product based on our favorite toy? Here we present a simple approach that allows such creative freedom. Using only 3-5 images of a user-provided concept, like an object or a style, we learn to represent it through new "wor
- background These Preprint. arXiv:2605.07257v1 [cs.CV] 8 May 2026 advances have fueled growing interest in generative personalization: adapting a pretrained T2I model to a user-specific concept (e.g., a person, pet, or object) from only a few reference images, while retaining the ability to place that concept into novel contexts via natural-language prompts [10, 30]. The core objective is to preserve the unique identity of the personal concept while remaining faithful to the prompt's semantics. Despite rapi
- background tasks including image synthesis [4, 24, 28], 3D object gen- eration [16, 21], and video production [1, 11, 29]. Leverag- ing large-scale pre-training on massive datasets, these mod- els now outperform earlier approaches in producing high- fidelity and coherent generative content. Current approaches range from slow fine-tuning methods like DreamBooth [26] and Textual Inversion [6], to zero- shot ID injection with encoders like IP-Adapter [38], Pho- toMaker [15], and InstantID [36], but these sacr
- background they frequently incur information loss in either foreground objects or background contexts. 2.2 Testing-Time Finetuning Testing-time finetuning methods constitute a fundamental para- digm for personalized image generation, where pre-trained model parameters are adaptively optimized for specific target subjects dur- ing inference to achieve high-fidelity customized image synthesis. Textual Inversion [11] first introduced the concept of optimizing the embeddings of learnable tokens by incorporatin
- background Finally, the model is highly sensitive to the prompt, and small changes in wording can lead to drastically different generated images, while semantically equivalent prompts may yield very different visual outputs [9,29]. Recently, a substantial body of works has tackled prompt inversion through optimization in continuous embedding or latent spaces [11,30,36,46]. While these methods can achieve high-fidelity reconstruction, they suffer from several fun- damental limitations. First, they assume wh
- background in the scene, while camera motion adjusts the camera's position and angle. 5.2.1 Motion Customization. Motion customization generates videos with motions matching reference videos, requiring disentanglement of motion and appearance. Customize-A-Video [210] utilizes Temporal LoRA (T-LoRA) to learn motion from temporal layers and Appearance Absorbers(e.g., spatial LoRA or textual inversion [211]) to isolate spatial features. MotionDi- rector [212] employs dual-path LoRA: spatial LoRAs capture appe
- background articulate the desired target through text descriptions. For instance, it is difficult to describe the precise features of an innovative toy car which is not encountered during large-scale model training. Consequently, the objective of customized generation is to enable the model to grasp new concepts from a minimal set of user-supplied images. Textual Inversion [243] addresses this by finding a new pseudo-word S˚ (similar to soft prompt discussed in Section III-A2) that represents new, specific
co-cited works
roles
background 15representative citing papers
PAPT uses adversarial prompt tuning on diffusion models to generate domain-style images while preserving category features, claiming superior single-domain generalization performance.
ChronoLock adds bounded perturbations to videos that target temporal denoising trajectories in T2V models, reducing unauthorized motion personalization on UCF Sports and HMDB51.
Introduces PexelsCustom-1M dataset, CustoMDiT parameter-efficient model, and OpenCustom benchmark for open-domain customized video generation.
ZIPP conditions diffusion models on LLM-rewritten prompts derived from graph-mined natural-language personas to achieve zero-shot personalization, reporting 13-20% gains and 79% human preference win rate over generic outputs.
LoRA-Key creates a standalone user-specific Watermark LoRA trained with a latent watermark prior and GOP, attachable via training-free superposition to protect LoRA ownership while preserving quality.
Orthogonal Negative Guidance subtracts only the orthogonal component of negative-prompt attention features from positive ones in FLUX models to suppress concepts while preserving semantics and quality.
PIU suppresses target identity generation in Arc2Face by replacing it with a proximity-selected anchor identity through localized fine-tuning of cross-attention layers while preserving output quality for other identities.
Tiny-Engram uses small n-gram-indexed memory tables to bind trigger phrases to target visual identities in diffusion models while preserving compositional control from the surrounding prompt.
Object functionalization is cast as neural graph completion over a functional graph of parts, contacts, and motions, followed by geometry realization that also rectifies erroneous motions, demonstrated on furniture with a new paired dataset.
Creativity is defined as meta-learning where a frozen diffusion creator optimizes candidates for rapid improvement by an adapting appraiser such as an autoencoder or CLIP adapter.
A training-free adaptive subspace projection method mitigates semantic collapsing in generative personalization by isolating and adjusting drift in a low-dimensional subspace using the stable pre-trained embedding as anchor.
Presents the first large-scale benchmark for multi-frame geometric distortion removal in videos under severe refractive warping, using real and synthetic data across four distortion levels and evaluating classical and learning-based methods including a proposed diffusion-based V-cache.
ASTRA disentangles subject identity from pose structure in diffusion transformers via retrieval-augmented pose guidance, asymmetric EURoPE embeddings, and a DSM adapter to improve multi-subject generation.
A 300K quadruplet dataset and UniDG foundation model enable reference- or text-driven defect generation across categories, outperforming few-shot baselines on anomaly detection tasks.
A coarse-to-fine pipeline deforms 3D meshes to reflect geometric features from an image using diffusion model representations while preserving topology and part-level semantics.
PAMELA provides a multi-user rating dataset and personalized reward model that predicts individual image preferences more accurately than prior population-level aesthetic models.
Training-free Riemannian fusion merges orthogonal style and concept adapters for diffusion models via geodesic approximation on GS matrices plus spectra restoration.
PromptEvolver recovers high-fidelity natural language prompts for given images by evolving them via genetic algorithm guided by a vision-language model, outperforming prior methods on benchmarks.
LPNSR derives optimal intermediate noise for diffusion SR via MLE and implements it with an LR-guided noise predictor, reaching SOTA perceptual quality in 4 steps without text priors.
A new data pipeline using real photos, entity removal, and image-to-video models plus a cross-view attention loss enables text-driven generation of actors in reference scenes with improved alignment.
MAGIC is a few-shot mask-guided anomaly inpainting framework using Gaussian prompt perturbation, spatially adaptive guidance, and context-aware mask alignment to produce high-fidelity, diverse anomalies that outperform prior methods on downstream detection tasks.
UniEdit-Flow presents tuning-free Uni-Inv and Uni-Edit methods for inversion and editing in flow models that achieve accurate reconstruction and robust region-preserving edits across generative models.
UniSim learns a universal real-world simulator from orchestrated diverse datasets, enabling zero-shot deployment of policies trained purely in simulation.
citing papers explorer
-
FLUX.1 Kontext: Flow Matching for In-Context Image Generation and Editing in Latent Space
FLUX.1 Kontext unifies image generation and editing via flow matching and sequence concatenation, delivering improved multi-turn consistency and speed on the new KontextBench benchmark.