flux.1-fill-dev-OneReward，Perfect Upgrade

• Edit workflow：http://i71i.com/mby8

• Register for free and get 1000 points. Log in every day and get points

**Perfect Upgrade ，**Seamlessly replace the fill model

OneReward is a novel visual domain RLHF approach that significantly improves the generative capabilities of strategy models across multiple subtasks by using Qwen2.5-VL as a generative reward model to enhance multi-task reinforcement learning. Based on OneReward, FLUX.1-Fill-dev-OneReward - Based on FLUX Fill [dev], it surpasses the closed-source FLUX Fill [Pro] in image restoration and epitaxy tasks, providing a powerful new benchmark for future unified image editing research.

------------------------------------------------

Base model:

black-forest-labs/ : Black Forest Labs.

bytedance-research/OneReward: The OneReward model developed by the ByteDance research team for reinforcement learning optimization.

yichengup/flux.1-fill-dev-OneReward: A model developed by yichengup that combines Flux.1-Fill-dev and OneReward, focusing on image filling and scaling tasks.

Label:

flux : Flux series models.

flux-fill : The image fill feature in the Flux series.

onereward: OneReward reinforcement learning methodology.

------------------------------------------------

FLUX.1-Fill-dev is an open-source image restoration and scaling model developed by Black Forest Labs, and the following is its detailed description:

Basic information

Model Architecture: Adopts the Rectified Flow Transformer architecture, combined with the generation capabilities of diffusion models, to intelligently fill in missing areas of images based on text prompts.

Parameter scale: 12 billion parameters.

Training method: Guidance distillation is used to optimize inference speed.

Licensing method: Model weights are publicly available and generated content can be used for personal, scientific, and commercial use, subject to the FLUX.1 [dev] Non-Commercial License.

Core features:

Image restoration: It can fill in missing or removed areas in the image based on text descriptions and binary masks, achieving high-precision image restoration.

Image Expansion: Support for outpainting, which seamlessly expands the boundaries of existing images.

Text Understanding and Generation: Understand complex text instructions and combine them with image context to generate natural, coherent restoration results.