Pruna AI releases LoRA for object removal from photos

Pruna AI has published P-Q-Image-2.1-Remove-Object LoRA adapters for the Qwen-Image-2.1 model, which remove objects from photos along with their shadows and reflections while preserving the remaining pixels, available in 6, 8, and 40-step variants.

Compiled by Sergey KostenchukPublished 2026-11-13Updated 2026-11-13

2026-11-13 Multimodal HuggingFace

🎨 Pruna AI releases LoRA for object removal from photos

The P-Q-Image-2.1-Remove-Object adapter set for Qwen-Image-2.1 erases objects along with their shadows and reflections without touching the other pixels. There is distillation for 6 and 8 steps (without CFG) and a base version for 40 steps; the 6-step version is recommended. The object is marked with a red outline or named via text.

🌍 Distillation compresses 40 steps down to 6-8 in the narrow task of inpainting, not across the entire generation pipeline; without CFG, a step costs approximately half as much. The weights are open: photo cleanup can be integrated into your own pipelines without an API.

👤 In ComfyUI or diffusers, the LoRA can be downloaded for free on Hugging Face. There is a ready-made workflow and a demo at dashboard.pruna.ai/playground. With a rough mask, the model may remove a neighboring item, and the filled area comes out smoother, with a mesh-like structure.

Source 1: https://huggingface.co/PrunaAI/P-Q-Image-2.1-Remove-Object Source 2: https://huggingface.co/PrunaAI/P-Q-Image-2.1-Remove-Object-Comfy

Sources