prithivMLmods/Qwen-Image-2.1-Object-Remover-Bbox-Preview

Qwen-Image-2.1-Object-Remover-Bbox-Preview

Qwen-Image-2.1-Object-Remover-Bbox-Preview is a preview LoRA adapter for Qwen-Image-2.1, designed for object removal using bounding box guidance. The adapter removes unwanted objects from selected regions while attempting to preserve surrounding textures, lighting, shadows, perspective, and overall image consistency.

This is an experimental preview model. Results may vary depending on the image, object, bounding box, and surrounding context.

Model Details

• Base Model: Qwen/Qwen-Image-2.1
• Adapter: Qwen-Image-2.1-Object-Remover-Bbox-Preview
• Model Type: LoRA / Adapter
• Model Status: Preview / Experimental
• Created by: prithivMLmods

Training Specifications

Parameter	Configuration
Dataset	50 pairs of high-quality images with bounding box annotations and manually manipulated resultant images
Save Precision	BF16
Learning Rate	1e-4
Optimizer	AdamW
Network Dimension (Rank)	16
Total Steps	5000
Trigger Prompt	Remove the red highlighted object from the scene

Download

Download the model files from the Files & versions tab:

Download Qwen-Image-2.1-Object-Remover-Bbox-Preview

Usage

Load the LoRA adapter with Qwen-Image-2.1 and provide an image with the target object indicated by a bounding box.

Use the trigger prompt:

Remove the red highlighted object from the scene

For best results, use a clear bounding box around the object to be removed.

Some Example Comparisons

Input	- LoRA (Base Model)	+ LoRA	Description & Settings
			Description: Removes unmarked objects and struggles at more than two objects (first silver and last marked black were not removed). Total Steps: 40
			Description: Base Model without LoRA struggles removing all bounding boxes (one cat remains unremoved). Total Steps: 40
			Description: Base Model without LoRA struggles to remove more than one object bounding box scenario. Total Steps: 40
			Description: Sometimes Base Model without LoRA struggles in retaining shadows after removal (look at the player's head shadow in the no-LoRA edition). Total Steps: 40

Limitations

This is an early preview release and may produce artifacts or inconsistencies in challenging cases, including:

• Complex backgrounds
• Large or overlapping objects
• Fine structures and textures
• Reflections and transparent objects
• Difficult lighting and perspective conditions

Results can vary depending on the input image and bounding box placement.

License

Please refer to the license terms of the base model, Qwen-Image-2.1, and ensure compliance with its terms when using or redistributing this adapter.

Acknowledgements

This adapter was created by prithivMLmods and is built for use with Qwen-Image-2.1.