README.md

---
title: Qwen-Image-Edit-2511-LoRAs-Fast
emoji: 🎃
colorFrom: blue
colorTo: blue
sdk: gradio
sdk_version: 6.20.0
app_file: app.py
pinned: true
license: apache-2.0
short_description: Demo of the Collection of Qwen Image Edit LoRAs
---

Prompt-based image editing on **Qwen-Image-Edit-2511**, with identity preservation (change pose/clothes/scene while keeping the person's face), HD output, and an anti-deformed-hands option.

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

---

# API

The Space exposes one endpoint: **`/edit`**. Base URL / Space id: `kulkas2pintu/QWEN_EDIT_IMAGE`.

### Parameters

| name | type | default | meaning |
|---|---|---|---|
| `image` | image (filepath/URL) | — | **required.** The image to edit (the person/subject whose identity is kept). |
| `prompt` | str | — | **required.** What to change, e.g. `"change her outfit to a red dress; turn to look left"`. |
| `lora_adapter` | str | `""` | Optional style LoRA name (see list below). `""` = no style / base edit. |
| `image2` | image | `None` | Optional 2nd reference image (e.g. a style or lighting reference). |
| `seed` | int | `0` | Seed; ignored when `randomize_seed=True`. |
| `randomize_seed` | bool | `True` | Pick a new random seed each call. |
| `guidance_scale` | float | `1.0` | true-CFG. `1.0` = fastest. `>1` (e.g. `2.5`) activates the negative prompt that suppresses extra fingers / identity drift (slower, ~2×). |
| `steps` | int | `4` | Denoise steps. `4` is native for this distilled model. |
| `preserve_identity` | bool | `True` | Keep the person's face while pose/head can change. |
| `identity_strength` | int | `65` | `0-100`, how forcefully identity is locked. |
| `output_size` | int | `1280` | Long-side pixels `1024-2048`. Higher = more HD detail, slower (~quadratic). |
| `fix_hands` | bool | `False` | Raise CFG/steps to suppress extra/fused fingers (slower). |

**Returns:** `[result_image, seed_used]`.

### Available `lora_adapter` names

`Multiple-Angles`, `Fal-Multiple-Angles` (best for pose/angle changes), `Photo-to-Anime`, `Anime-V2`,
`Light-Migration`, `Any-light`, `Studio-DeLight`, `Cinematic-FlatLog`, `Upscaler`, `Unblur-Anything`,
`Style-Transfer`, `Manga-Tone`, `Anything2Real`, `Polaroid-Photo`, `Midnight-Noir-Eyes-Spotlight`,
`Hyper-Realistic-Portrait`, `Ultra-Realistic-Portrait`, `Pixar-Inspired-3D`, `Noir-Comic-Book`.

> Tip: for realistic edits keep `preserve_identity=True`. **Uncheck it** (`preserve_identity=False`)
> for full-transform styles like `Photo-to-Anime` / `Pixar-Inspired-3D`.

## Python (`gradio_client`)

```bash
pip install gradio_client
```

```python
from gradio_client import Client, handle_file

client = Client("kulkas2pintu/QWEN_EDIT_IMAGE")  # add hf_token="hf_..." for a private Space

result_image, seed_used = client.predict(
    image=handle_file("https://example.com/person.jpg"),  # or a local path
    prompt="change her outfit to an elegant red dress; turn her head to the left",
    lora_adapter="",            # or e.g. "Fal-Multiple-Angles" for pose changes
    image2=None,
    seed=0,
    randomize_seed=True,
    guidance_scale=2.5,          # >1 turns on hand/identity negatives
    steps=6,
    preserve_identity=True,
    identity_strength=80,
    output_size=1536,            # HD
    fix_hands=True,
    api_name="/edit",
)
print("saved to:", result_image, "| seed:", seed_used)
```

`result_image` is a local filepath to the PNG result.

## JavaScript (`@gradio/client`)

```bash
npm i @gradio/client
```

```js
import { Client } from "@gradio/client";

const image = await (await fetch("https://example.com/person.jpg")).blob();
const client = await Client.connect("kulkas2pintu/QWEN_EDIT_IMAGE");

const res = await client.predict("/edit", {
  image,
  prompt: "change her outfit to a red dress",
  lora_adapter: "",
  guidance_scale: 2.5,
  steps: 6,
  output_size: 1536,
  fix_hands: true,
});
console.log(res.data);   // [result_image, seed_used]
```

## curl (raw REST)

Gradio's REST API is a two-step call (queue → result):

```bash
# 1) submit the job (image can be a public URL)
curl -s -X POST https://kulkas2pintu-qwen-edit-image.hf.space/gradio_api/call/edit \
  -H "Content-Type: application/json" \
  -d '{"data":[
        {"path":"https://example.com/person.jpg"},
        "change her outfit to a red dress",
        "", null, 0, true, 2.5, 6, true, 80, 1536, true
      ]}'
# -> {"event_id":"<id>"}

# 2) stream the result
curl -N https://kulkas2pintu-qwen-edit-image.hf.space/gradio_api/call/edit/<id>
```

The `data` array is positional, in the same order as the parameter table above.

> The live, always-up-to-date schema is at
> `https://kulkas2pintu-qwen-edit-image.hf.space/gradio_api/` (the "view API" JSON).
> This Space also runs as an **MCP server** (`/gradio_api/mcp/`), so `/edit` is usable
> as a tool from MCP-compatible clients.