PosterCopilot - A poster design mockup jointly launched by Nanjing University and LibLib.ai, among others.

What is PosterCopilot?

PosterCopilot is a professional-grade poster design model jointly developed by Nanjing University, LibLib.ai, and the Institute of Automation, Chinese Academy of Sciences. Through a unique three-stage training strategy, the model possesses powerful layout reasoning and precise editing capabilities, enabling end-to-end design from material planning to the final draft. The model supports full material generation, missing material completion, and multi-round refined editing, and is equipped with a high-quality hierarchical poster dataset. It addresses the geometric, visual, and editing pain points of existing design models, providing a new paradigm for AI-assisted creative design.

PosterCopilot's main functions

• Full-material poster generationWhen users provide complete materials, the model can accurately arrange multimodal elements to generate professional posters that conform to aesthetic principles, while ensuring zero distortion of the materials.
• Intelligent material completionWhen materials are missing, it automatically synthesizes a background or foreground decorative layer with a unified style, achieving a seamless transformation from abstract ideas to a complete poster.
• Multiple rounds of refined editing:
  • Precise single-layer editing: Modify only specific layers (such as changing the model's hair color or changing the object's material), while leaving other areas unchanged.
  • Global topic migrationOne-click switching of poster themes, automatic replacement of the main subject and adjustment of related elements, while retaining the original layout.
  • Intelligent size reconstructionThe layout is re-inferred based on the canvas size parameters to adapt to different media layouts.
• Creative transformationIt supports seamless transformation from abstract design concepts to specific materials, generating high-quality engineering-grade prompts and driving the generation of materials that match the style.

PosterCopilot's technical principles

• Progressive three-stage training strategy:
  • Phase 1: Perturbation-monitored fine-tuning (PSFT)Introducing Gaussian noise perturbation allows the model to learn coordinate distribution rather than discrete points, correcting the distortion of geometric space and improving the geometric rationality of the layout.
  • Phase Two: Visual-Reality Alignment Reinforcement Learning (RL-VRA): Introducing a reward signal based on DIoU and element fidelity to correct overlap and proportion issues, ensuring the visual realism of the layout.
  • Phase 3: Aesthetic Feedback Reinforcement Learning (RLAF)The model uses aesthetic rewards to encourage it to generate more visually impactful and diverse layouts, going beyond a single truth value.
• Generative Agent:By combining the Reception Model and the T2I Model, a seamless transformation from abstract inspiration to concrete content is achieved. The Reception Model breaks down user intent into detailed plans, generates engineering-grade prompts, and drives the T2I Model to generate high-quality content.
• High-quality hierarchical poster dataset:We constructed a dataset containing 160,000 professional posters and a total of 2.6 million layers. By using OCR-assisted fine-grained layer fusion technology, we solved the problem of excessive layer fragmentation in traditional datasets, providing rich and high-quality data support for model training.

PosterCopilot project address

• Project official websitehttps://postercopilot.github.io/
• GitHub repositoryhttps://github.com/JiazheWei/PosterCopilot
• arXiv technical paper: https://arxiv.org/pdf/2512.04082

Application scenarios of PosterCopilot

• Advertising and MarketingQuickly generate posters that match your brand and marketing themes, adapt to multiple platform sizes, support multiple rounds of modifications, and meet your advertising needs.
• Creative Design StudioAs a designer's aid, it can quickly generate preliminary plans, support material completion and optimization, and improve design efficiency and creative inspiration.
• In-house design teamStandardized design processes enable rapid response to market changes, support multi-departmental collaboration, and improve overall team efficiency.
• Education and TrainingAs a teaching tool, it helps students understand design principles, stimulates creativity, and is suitable for design teaching and creative inspiration scenarios.
• Cultural and artistic institutionsDesign artistic posters for exhibitions and cultural events, and support the generation of design schemes that are consistent with the cultural atmosphere.