flux.1-fill-dev-OneReward-int8_convrot

经过测试，相比于GGUF Q5，采样速度加快了100%，200万像素的图像采样只需要20秒（rtx3080-20g），生成质量方面相比gguf更佳

只能使用ComfyUI-INT8-Fast的插件节点加载

相比Qwen edit，flux2等图像编辑模型，它的优点是速度更快，颜色一致性保持更佳和边缘修复自然，无需提示词完成遮罩内的图像瑕疵修复，完美的照片修复模型

模型介绍：

这是一个统一的、目前最先进的图像重绘修复模型，能够高效处理包括图像填充、图像扩展、物体移除和文本渲染在内的多种任务。它超越了包括 Ideogram、Adobe Photoshop 和 FLUX Fill [Pro] 在内的多个领先的商业和开源系统。最后，基于 FLUX Fill [dev]，发布了 FLUX.1-Fill-dev-OneReward ，它在图像修复和图像扩展任务中均优于闭源的 FLUX Fill [Pro]，为未来统一图像编辑领域的研究提供了一个强大的新基准。

Testing shows that, compared with GGUF Q5, sampling speed is improved by 100%. Sampling a 2-megapixel image takes only 20 seconds（rtx3080-20g）, and the output quality is also better than GGUF.

It can only be loaded using the plugin nodes from ComfyUI-INT8-Fast.

Compared with image editing models such as Qwen Edit and FLUX.2, its advantages are faster speed, better color consistency, and more natural edge restoration. It can repair image defects within a masked area without requiring prompts, making it an excellent photo restoration model.

Model Overview:

This is a unified, state-of-the-art image inpainting and restoration model that can efficiently handle a wide range of tasks, including image inpainting, outpainting, object removal, and text rendering. It outperforms several leading commercial and open-source systems, including Ideogram, Adobe Photoshop, and FLUX Fill [Pro]. Finally, based on FLUX Fill [dev], FLUX.1-Fill-dev-OneReward was released. It surpasses the closed-source FLUX Fill [Pro] in both image inpainting and outpainting tasks, providing a strong new benchmark for future research in unified image editing.