New model support

### Feature Idea

https://github.com/bytedance/OneReward
A new visual domain RLHF method called OneReward, which employs Qwen2.5-VL as a generative reward model to enhance multi-task reinforcement learning, significantly improves the generative ability of the policy model across multiple subtasks. Building on OneReward, we have developed Seedream 3.0 Fill, a unified SOTA image editing model that can effectively handle various tasks, including image filling, image expansion, object removal, and text rendering. It outperforms several leading commercial and open-source systems, including Ideogram, Adobe Photoshop, and FLUX Fill [Pro].

### Existing Solutions

_No response_

### Other

_No response_

The issue names OneReward and Seedream 3.0 Fill but provides no repository file, test, or entry point. Start by reviewing how ComfyUI adds model support and clarifying the required integration scope; done would mean defined, working support for the requested model.

Written by the indexing model from the issue text.