# Image Quality

SeedEdit is a large diffusion model introduced by the Doubao Team, designed to revise images based on any text prompts. It achieves an optimal balance between image reconstruction and regeneration by aligning a powerful image generator with a robust image editor in a stepwise fashion. SeedEdit enables zero-shot stable editing of high aesthetic/resolution images and supports continuous revision of images. The significance of this technology lies in its ability to address the core challenge of scarcity of paired image data in image editing problems, by viewing the text-to-image (T2I) generation model as a weak editing model and achieving 'editing' by generating new images with new prompts, which are then distilled and aligned within an image-conditioned editing model.

PanoFree is an innovative technology for generating panoramic multi-view images that addresses issues of consistency and cumulative error through iterative deformation and coloring without the need for additional tuning. Experimental results demonstrate a significant reduction in errors, improved global consistency, and enhanced image quality without the complexity of manual adjustments. Compared to existing methods, PanoFree is more efficient in terms of time and GPU memory usage, while maintaining diversity in results.

Clipdrop Image upscaler is a handy tool that can improve the quality of compressed images. Its main advantages include fast image quality enhancement, a variety of image processing functionalities, and ease of use. Background information includes the product's positioning as a tool for improving image quality.

UniFL is a project designed to enhance the quality of generative models and accelerate inference speeds. It effectively addresses the existing issues of image quality, aesthetic appeal, and inference speed in current diffusion models through three key components: perceptual feedback learning, decoupled feedback learning, and adversarial feedback learning. Through experimental validation and user studies, UniFL has demonstrated significant performance improvements and strong generalization capabilities across multiple diffusion models.

TextCraftor is an innovative text encoder fine-tuning technique that can significantly improve the performance of text-to-image generation models. Through reward function optimization, it enhances image quality and text alignment without requiring additional datasets.

StreamingT2V is an advanced autoregressive technique that creates long videos with rich, dynamic motion without any stuttering. It ensures temporal consistency within the video, aligning closely with the descriptive text, and maintains high-frame-rate image quality.

This paper proposes a new metric for evaluating image generation models. It highlights the issues with the Frechet Inception Distance (FID) metric and introduces a new metric called CMMD. Extensive experiments demonstrate that the FID metric may be unreliable for evaluating text-to-image models, while the CMMD metric can more reliably assess image quality.