Blame for README.md

---
license: other
license_name: qwen-research
license_link: https://huggingface.co/Qwen/Qwen-Image-2.1/blob/main/LICENSE
base_model: PrunaAI/P-Q-Image-2.1-Remove-Object
pipeline_tag: image-to-image
tags:
  - comfyui
  - qwen
  - qwen-image
  - image-editing
  - object-removal
  - inpainting
  - lora
  - few-step
  - pruna-ai
---

<!-- header start -->
<p align="center">
  <img src="assets/P-Q-Image-2.1-Remove-Object.png" width="800">
</p>
<!-- header end -->

<p align="center">
  <a href="https://github.com/PrunaAI/pruna">
    <img src="https://img.shields.io/badge/GitHub-PrunaAI-9334E9?style=plastic&logo=github&logoColor=white" alt="GitHub">
  </a>
  &nbsp;
  <a href="https://twitter.com/PrunaAI">
    <img src="https://img.shields.io/badge/Twitter%2FX-@PrunaAI-9334E9?style=plastic&logo=x&logoColor=white" alt="Twitter/X">
  </a>
  &nbsp;
  <a href="https://www.linkedin.com/company/pruna-ai">
    <img src="https://img.shields.io/badge/LinkedIn-PrunaAI-9334E9?style=plastic&logo=linkedin&logoColor=white" alt="LinkedIn">
  </a>
  &nbsp;
  <a href="https://discord.com/invite/JFQmtFKCjd">
    <img src="https://img.shields.io/badge/Discord-Join%20us-9334E9?style=plastic&logo=discord&logoColor=white" alt="Discord">
  </a>
</p>

<div align="center">

<h1 style="color: #9334E9;">⚡ P-Q-Image-2.1-Remove-Object</h1>

<h2>Object removal LoRAs for Qwen-Image-2.1 that leave the rest of the image untouched — ComfyUI</h2>

<h3>
  <span style="color: #9334E9;">6 or 8 steps</span>
  &nbsp;·&nbsp;
  <span style="color: #9334E9;">No CFG</span>
  &nbsp;·&nbsp;
  <span style="color: #9334E9;">Mask optional</span>
  &nbsp;·&nbsp;
  Ready-to-use workflows
</h3>

</div>

**Built with Qwen.**

**P-Q-Image-2.1-Remove-Object** is a set of LoRA adapters for
[`Qwen/Qwen-Image-2.1`](https://huggingface.co/Qwen/Qwen-Image-2.1) that **remove the object you point
at, and change nothing else**. Mark the object with a mask (painted in ComfyUI's MaskEditor or loaded
from a file), write what to remove, and the object is replaced by what would be behind it. The rest of
the image stays as it was.

This repository is the ComfyUI export of
[`PrunaAI/P-Q-Image-2.1-Remove-Object`](https://huggingface.co/PrunaAI/P-Q-Image-2.1-Remove-Object),
with ready-to-use workflows, an example and a small custom node. The few-step variants run on top of
PrunaAI's distillation LoRAs from
[`PrunaAI/P-Q-Image-2.1-Comfy`](https://huggingface.co/PrunaAI/P-Q-Image-2.1-Comfy).

<p align="center">
  <img src="examples/p_q_image_2.1_remove_object_example_before_after.jpg" width="800">
</p>

*"Remove the phone and the raised arm holding it."* The example the workflows open with, 6-step
workflow as shipped. Photo: Aliaksei Antropau,
[Unsplash](https://unsplash.com/photos/person-recording-concert-under-red-lights-n24LASzjRgs).

## Variants

Pick **6-step** as the default (recommended), or **base** to avoid the few-step model's drift.

| Variant | File | Distillation LoRA (`PrunaAI/P-Q-Image-2.1-Comfy`) | Steps | Strength | Workflow |
|---|---|---|:---:|:---:|---|
| `6-step` | `p_q_image_2.1_remove_object_6_steps_v0.1_comfyui_bf16.safetensors` | `p_q_image_2.1_6_steps_v0.2_comfyui_bf16.safetensors` | 6 | **0.7** | `workflows/p_q_image_2.1_remove_object_v0.1_6step.json` |
| `8-step` | `p_q_image_2.1_remove_object_8_steps_v0.1_comfyui_bf16.safetensors` | `p_q_image_2.1_8_steps_v0.1_comfyui_bf16.safetensors` | 8 | **0.7** | `workflows/p_q_image_2.1_remove_object_v0.1_8step.json` |
| `base` | `p_q_image_2.1_remove_object_base_v0.1_comfyui_bf16.safetensors` | none (base model alone) | 40 | 1.0 | none |

"Strength" is the adapter weight of this LoRA; the distillation LoRA always runs at 1.0. The few-step
variants are made to run on top of the matching distillation LoRA, with its own step count and sigmas,
without CFG: the workflows set all of this. The weights are those of
`PrunaAI/P-Q-Image-2.1-Remove-Object`, converted to ComfyUI's LoRA format (bf16).

## Quickstart

1. **Update ComfyUI** to a version with native Qwen-Image-2.1 support.
2. **Download the models** from [Comfy-Org/Qwen-Image-2.1](https://huggingface.co/Comfy-Org/Qwen-Image-2.1),
   [PrunaAI/P-Q-Image-2.1-Comfy](https://huggingface.co/PrunaAI/P-Q-Image-2.1-Comfy) and this repo
   (each workflow's *Model links* note has the direct links):

   ```
   📂 ComfyUI/
   ├── 📂 custom_nodes/
   │   └── 📂 precise_removal_outline/                  (this repo)
   ├── 📂 input/
   │   └── p_q_image_2.1_remove_object_example.png      (this repo, examples/)
   └── 📂 models/
       ├── 📂 diffusion_models/
       │   └── qwen_image_2.1_int8_convrot.safetensors
       ├── 📂 loras/
       │   ├── p_q_image_2.1_6_steps_v0.2_comfyui_bf16.safetensors
       │   └── p_q_image_2.1_remove_object_6_steps_v0.1_comfyui_bf16.safetensors   (this repo)
       ├── 📂 text_encoders/
       │   ├── qwen3vl_8b_int8_convrot.safetensors
       │   └── qwen3.5_9b_qwen_image_2.1_pe_i2i.int8_convrot.safetensors
       └── 📂 vae/
           └── qwen_image_2.1_vae_bf16.safetensors
   ```

   For the 8-step variant, use the two 8-step LoRAs instead. The prompt-enhancer model
   (`qwen3.5_9b…`) is switched off in the workflows, but ComfyUI checks that it is present.
3. **Install the custom node**: copy `precise_removal_outline/` into `ComfyUI/custom_nodes/`, install its
   `requirements.txt` (opencv) into ComfyUI's Python and restart ComfyUI. It adds the *Precise Removal
   Outline* node (category Pruna).
4. **Open a workflow**: drag `workflows/p_q_image_2.1_remove_object_v0.1_6step.json` (or the 8-step one)
   onto ComfyUI and press **Run**. It opens on the example above, with the mask already painted.

## Using the workflow

- **Load Image**: your photo.
- **Mask (optional)**: right-click Load Image, *Open in MaskEditor*, paint over the object, *Save*. Or
  load a mask file: select the greyed-out *Load Mask from file* node, press Ctrl+B (Cmd+B on Mac) to
  switch it on, and pick the file (white = the object, black = keep, any size). With both, they are
  combined. Without a mask, the instruction alone says what to remove.
- **Precise Removal Outline**: the instruction, e.g. *"Remove the man on the left."* With a mask, the
  node draws the outline exactly as expected by the model and adds the sentence that goes with it
  (*"The area to edit is outlined in red. Remove the red outline as well."*).
- **Run**. Image Compare shows your photo against the result (its slider needs ComfyUI's Nodes 2.0).

The edit box shows only `seed` and `removal_strength`; everything else is fixed inside it (double-click
to see). The result has the photo's exact size: the model generates at the nearest multiples of 32 and
the result is resized back.

### Recommended settings

- **The mask does not need to be exact.** Rough, generous masks work. Include the object's shadow or
  reflection if they should go too, but keep the mask off anything that should stay.
- **Name the object as it appears**, and say which one when there are several similar ones (*"the man
  on the right"*).
- **Keep `removal_strength` at 0.7.** Lower strengths reduce texture artifacts, but at 0.5 some removals
  fail (the object stays, recolored); much higher, the model also removes neighboring objects.
- **Seed**: fixed. If a removal leaves something behind, try another seed.

## Limitations

- **It can remove more than the outlined object**, when the mask also covers part of something that
  should stay. Keep the mask tight where other things overlap the object.
- **Textures in the filled area can look off**: smoother than their surroundings, or with a faint
  regular grid. Photos well under 1 MP also keep less fine texture outside the mask.
- **Only for removing objects.** Other edits are not expected to work better than with the base model.

## License

P-Q-Image-2.1-Remove-Object is a derivative of Qwen-Image-2.1 and follows the
[Qwen RESEARCH LICENSE AGREEMENT](https://huggingface.co/Qwen/Qwen-Image-2.1/blob/main/LICENSE) (copy in
`LICENSE`, attribution in `NOTICE`). Review its use restrictions before using or redistributing the
adapters. The few-step variants also need PrunaAI's distillation LoRA
([PrunaAI/P-Q-Image-2.1-Comfy](https://huggingface.co/PrunaAI/P-Q-Image-2.1-Comfy)), whose own terms
apply.

Qwen is licensed under the Qwen RESEARCH LICENSE AGREEMENT, Copyright (c) 2026 Hangzhou Tongyi
Laboratory Technology Co., Ltd. All Rights Reserved.

The example photo is from Unsplash, under the [Unsplash License](https://unsplash.com/license). The
Unsplash License does not cover people in a photo: the person in the example is seen from behind, in
shadow, without a readable face.

## What's next?

- **Use P-Q-Image-2.1-Remove-Object to remove objects from photos with Qwen-Image-2.1 in a few steps.**
- Compress your own models with [Pruna](https://github.com/PrunaAI/pruna) and give us a ⭐️ for more efficiency!

<style>
.model-button {
  display: inline-flex;
  flex-direction: row;
  justify-content: center;
  align-items: center;
  gap: 8px;

  padding: 8px 20px;
  border: none;
  border-radius: 8px;

  background: #9334e9;
  color: #ffffff;

  font-size: 14px;
  font-weight: 400;
  line-height: 1;
  text-decoration: none;
  white-space: nowrap;
  cursor: pointer;

  box-sizing: border-box;
  overflow: visible;
  opacity: 1;
}
</style>

<a href="https://dashboard.pruna.ai/login?utm_source=huggingface&utm_medium=model_card&utm_campaign=hf_traffic" class="model-button">Try our models</a>