cs.CVMar 6, 2026

Pinterest Canvas: Large-Scale Image Generation at Pinterest

Authors: Yu Wang, Eric Tzeng, Raymond Shiau, Jie Yang, Dmitry Kislyuk, Charles Rosenberg

Organizations: Pinterest, Inc. San Francisco, CA, USA

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

While recent image generation models demonstrate a remarkable ability to handle a wide variety of image generation tasks, this flexibility makes them hard to control via prompting or simple inference adaptation alone, rendering them unsuitable for use cases with strict product requirements. In this paper, we introduce Pinterest Canvas, our large-scale image generation system built to support image editing and enhancement use cases at Pinterest. Canvas is first trained on a diverse, multimodal dataset to produce a foundational diffusion model with broad image-editing capabilities. However, rather than relying on one generic model to handle every downstream task, we instead rapidly fine-tune variants of this base model on task-specific datasets, producing specialized models for individual use cases. We describe key components of Canvas and summarize our best practices for dataset curation, training, and inference. We also showcase task-specific variants through case studies on background enhancement and aspect-ratio outpainting, highlighting how we tackle their specific product requirements. Online A/B experiments demonstrate that our enhanced images receive a significant 18.0% and 12.5% engagement lift, respectively, and comparisons with human raters further validate that our models outperform third-party models on these tasks. Finally, we showcase other Canvas variants, including multi-image scene synthesis and image-to-video generation, demonstrating that our approach can generalize to a wide variety of potential downstream tasks.

Figures & tables

Figure 1. Rather than relying solely on a single generic image generation model to handle every potential task, in our design of Pinterest Canvas we opt to rapidly fine-tune dedicated variants that are tailored to each use case’s unique product requirements. A diagram showcasing our base model's ability to handle flexible yet generic image editing tasks, as well as three example task-specific variants: image enhancement for advertising, multi-image scene synthesis, and image-to-motion generation.
Figure 2. Multimodal dataset visualization across tasks. In each example, the left image(s) is the input condition(s), and the right image is the target output; the text below each pair describes the transformation required to produce the target from the input.
Figure 3. Timestep shifting experiments. Averaged EditReward scores and qualitative examples at each m/n\sqrt{m/n} setting. Timestep shifting
Figure 4. Canvas outpainting inference pipeline. Canvas outpainting inference pipeline diagram
Figure 5. Outpainting VAE comparison. Outpainting VAE
Figure 6. Visual examples for background outpainting. Visual examples for background outpainting
Figure 7. Visual examples for aspect-ratio outpainting. Visual examples for aspect ratio outpainting
Figure 8. Failure cases of third-party models. Failure Cases of Third-party Models
 No-defect Rate
ModelProductBackgroundOverall
GPT-Image 1.552.9%50.8%26.2%
FLUX.1 Kontext53.4%55.8%28.2%
Nano Banana74.6%55.9%42.5%
Canvas84.0%54.9%47.2%
Table 1. Human evaluation for background outpainting.
MetricBackgroundAspect-ratio
CTR+18.0%+12.5%
gCTR30+7.6%+6.8%
Click Volume+18.6%+12.9%
Table 2. Engagement results for outpainting use-cases.
Figure 9. Visual examples for multi-image scene synthesis. Visual examples for multi-image scene synthesis
Figure 10. Visual examples for image-to-motion generation. Visual examples for image-to-motion generation
Appendix figures & tables3 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 11. Human review questions for background outpainting. Human review questions for background outpainting.
Figure 12. Human review questions for aspect-ratio outpainting Human review questions for aspect-ratio outpainting
Figure 13. Visualizations of more product editing examples More product use-case examples
More in this topic

Image Generation

Image generation studies models that create or edit images from text, examples, layouts, and other structured controls.

See all 463 papers

Explore similar work

CardsList
Jun 1, 2026

Polaris: Scaling Up Instruction-Guided Image Generation Towards Millions of Personalized Style Needs

Users increasingly expect image generation models to quickly adapt to highly diverse and personalized requirements, such as producing images with distinctive styles or characteristics. Traditional approaches rely on fine-tuning, which is costly and difficult to scale. To cope with these limitations, the community has accumulated a growing library of fine-tuned modules and adapters, where each component targets specific generation needs and collectively serves as a foundation for handling new demands. This naturally raises a question: instead of repeatedly training new models, can we systematically exploit this expanding ecosystem to better fulfill user instructions? To this end, we present Polaris, an intelligent retrieval framework that automatically selects and integrates suitable models from the model library based on a user's instructions. The key insight is that harnessing such a massive and heterogeneous pool requires not only finding the most relevant modules among thousands of candidates, but also aligning them effectively for instruction-driven generation and editing. Polaris addresses this challenge by indexing over 6,500 checkpoints and 75,000 adapters, and retrieving the most relevant components given a user's input and instruction. In doing so, it delivers scalable, controllable, and well-aligned generation -- without any additional training.

Zhi-Kai Chen, Jun-Peng Jiang, Jun-Jie Tao +2

School of Artificial Intelligence, Nanjing University, China · National Key Laboratory for Novel Software Technology, Nanjing University, China · Nanjing University, China

Image GenerationControllable Image Generation
AskPDF
Sep 30, 2026

Personalized Image Generation with Reasoning and Reflection

Personalized image generation has remained narrowly focused on conditional synthesis from curated visual exemplars, rather than capturing who a user is. In practice, however, a user's personal context is much richer, comprising reviews, posts, images, captions, and metadata accumulated over time. A truly personalized generator should leverage this history to produce images aligned with the user's lifestyle and aesthetic preferences. To this end, we introduce the first unified benchmark for personalized image generation from user histories. The benchmark comprises two complementary tasks and a multi-axis evaluation protocol that assesses target fidelity, visual quality, user distinguishability, semantic alignment with the user's history, and task-specific utility. Grounded in real-world e-commerce and social media settings, the benchmark includes: (1) Personalized Scene Generation, which places a given object in a scene that reflects a user's preferences and lifestyle, motivated by personalized product presentation; and (2) Personalized Creative Generation, which generates a novel image on a specified topic that is faithful to a user's aesthetic and visual identity, motivated by social media content creation. We further propose PEARL, which couples a multimodal reasoner with a frozen image generator in an interleaved reason-reflect loop optimized with differential data reward. Across both tasks, PEARL outperforms strong baselines, achieving an average improvement of 15% across personalization metrics.

Bo Ni, Ngoc N. Tran, Qinwen Ge +10

Vanderbilt University · Adobe Systems · University of Maryland, College Park +1

Image GenerationAgentic Image GenerationImage Generation EvaluationPersona-Conditioned Generation
AskPDF
May 29, 2026

APE: Agentic Prompt Enhancer for Image Generation and Editing

Natural language has become a powerful interface for image generation and editing, yet text-guided visual systems remain highly sensitive to prompt formulation. Semantically similar requests can produce different outputs depending on wording, specificity, and how explicitly visual constraints are stated, motivating prompt enhancement as a trainable component rather than a peripheral user choice. Existing strong enhancers often rely on large, proprietary LLMs such as ChatGPT or Gemini, adding cost, latency, and deployment dependence to the visual generation pipeline. We propose Agentic Prompt Enhancer (APE), a lightweight framework that post-trains small language models (SLMs) as prompt-enhancement agents. APE supports both single-agent rewriting and role-specialized multi-agent enhancement. Its single-agent instantiation, SAPE, rewrites the prompt in one pass, while its multi-agent instantiation, MAPE, decomposes enhancement into a router--rewriter--composer process for handling compositional constraints over objects, attributes, spatial relations, and edits. With task-aware rewards and post-training protocols, APE improves visual alignment and prompt following without modifying the downstream visual model. Experiments on challenging image generation and editing benchmarks demonstrate that post-trained small prompt enhancers reliably outperform their base counterparts, narrowing the gap to closed-source prompt enhancers; in addition, MAPE proves particularly strong on complex compositional tasks within these benchmarks.

Zijian Huang, Jay Zhangjie Wu, Zian Wang +5

1NVIDIA · University of Michigan

Text-Guided Image EditingT2I GenerationSmall Language ModelsCompositional T2I GenerationPrompt Optimization for T2I Generation
AskPDF