# Wan 2.5 (Preview)

> Alibaba's text-to-image and image-to-image generation model from the Wan AI suite, offering high-quality visual generation capabilities

## Quick Reference

- Model ID: wan-2.5-preview
- Creator: Alibaba
- Status: active
- Family: wan
- Base URL: https://api.lumenfall.ai/openai/v1

## Specifications

- Input Modalities: text, image
- Output Modalities: image
- Supported Modes: Text to Image, Image Edit

## API Parameters

The compiled parameter schema for this model is available via the API: `GET /v1/models/wan-2.5-preview?schema=true`.

### Core Parameters

- `prompt` (string) — REQUIRED: Text prompt for image generation. Modes: Text to Image, Image Edit
- `negative_prompt` (string): Negative prompt to guide generation away from undesired content. Modes: Text to Image, Image Edit
- `seed` (integer): Random seed for reproducibility. Modes: Text to Image, Image Edit

### Size & Layout

- `size` (string): Image dimensions as WxH pixels (e.g. "1024x1024") or aspect ratio (e.g. "16:9"). Modes: Text to Image, Image Edit
- `aspect_ratio` (string): Aspect ratio of the output image (e.g. "16:9", "1:1"). Modes: Text to Image, Image Edit
- `resolution` (string): Output resolution tier (e.g. "1K", "4K"). Modes: Text to Image, Image Edit

### Media Inputs

- `image` (file) — REQUIRED: Input image(s) to edit. Modes: Image Edit

### Output & Format

- `response_format` (string): How to return the image. Default: url. Values: url, b64_json. Modes: Text to Image, Image Edit
- `output_format` (string): Output image format. Values: png, jpeg, gif, webp, avif. Modes: Text to Image, Image Edit
- `output_compression` (integer): Compression level for lossy formats (JPEG, WebP, AVIF). Modes: Text to Image, Image Edit
- `n` (integer): Number of images to generate. Default: 1. Modes: Text to Image, Image Edit

### Additional Parameters

- `prompt_enhancement` (string): Whether an LLM rewrites/expands the prompt before generation (off, on). Values: off, on. Modes: Text to Image, Image Edit
- `enable_safety_checker` (boolean): If set to true, the safety checker will be enabled.. Modes: Text to Image, Image Edit. Only available via fal


## Model Identifiers

- Primary Slug: wan-2.5-preview

## Dates

- Released: September 2025

## Tags

image-generation, text-to-image, image-editing

## Available Providers

### fal.ai

- Config Key: fal/wan-2.5-preview
- Provider Model ID: fal-ai/wan-25-preview/text-to-image
- Pricing: $0.050/image
  - Source: https://fal.ai/models/fal-ai/wan-25-preview/text-to-image

### fal.ai

- Config Key: fal/wan-2.5-preview-edit
- Provider Model ID: fal-ai/wan-25-preview/image-to-image
- Pricing: $0.050/image
  - Source: https://fal.ai/models/fal-ai/wan-25-preview/text-to-image


## Arena Benchmarks

### Candid Street Photography

- Elo: 1181
- Record: 5W / 0L / 0T (5 battles)
- Rank: #18 of 68

### Chalkboard Menu

- Elo: 1079
- Record: 2W / 1L / 1T (4 battles)
- Rank: #28 of 69


## Use Cases & Category Performance

### Photorealism (Text-to-Image)

- Rank: #16 of 62
- Elo: 1205
- Record: 15W / 5L / 1T (21 battles)
- Win Rate: 71.4%

### Aesthetics (Text-to-Image)

- Rank: #22 of 62
- Elo: 1226
- Record: 14W / 5L / 1T (20 battles)
- Win Rate: 70.0%

### Prompt Adherence (Text-to-Image)

- Rank: #29 of 61
- Elo: 1222
- Record: 12W / 3L / 2T (17 battles)
- Win Rate: 70.6%

### Product, Branding & Commercial (Text-to-Image)

- Rank: #38 of 61
- Elo: 1185
- Record: 8W / 2L / 1T (11 battles)
- Win Rate: 72.7%

### Preservation (Image Editing)

- Rank: #24 of 33
- Elo: 1149
- Record: 4W / 3L / 0T (7 battles)
- Win Rate: 57.1%

### Text Rendering (Text-to-Image)

- Rank: #44 of 61
- Elo: 1171
- Record: 8W / 2L / 2T (12 battles)
- Win Rate: 66.7%

### Photorealism (Image Editing)

- Rank: #22 of 28
- Elo: 1144
- Record: 3W / 2L / 0T (5 battles)
- Win Rate: 60.0%

### Lighting, Shadows & Materials (Image Editing)

- Rank: #24 of 30
- Elo: 1146
- Record: 3W / 2L / 0T (5 battles)
- Win Rate: 60.0%

### Aesthetics (Image Editing)

- Rank: #16 of 17
- Elo: 1156
- Record: 3W / 1L / 0T (4 battles)
- Win Rate: 75.0%

### Creativity (Text-to-Image)

- Rank: #59 of 62
- Elo: 1095
- Record: 3W / 4L / 1T (8 battles)
- Win Rate: 37.5%


## Image Gallery

3 images available for this model. Browse all at https://lumenfall.ai/models/alibaba/wan-2.5-preview/gallery

### Curated Examples

- [A wide cinematic shot of a sophisticated, high-end design studio at sunset, with warm amber light...](https://assets.lumenfall.ai/SqtsAV8LGZBaSgJZbdFcDfNF1HUu4naDSIEllbs-a3Q/rs:fit:1500:1500/plain/gs://lumenfall-prod-assets/l5cmpz0n90fajeztbsrb9dkchz6c@jpeg)

### Arena Competition Results

- [Candid Street Photography](https://assets.lumenfall.ai/3Sexu1Ub5oVn_3DTnIFI3p4pAj_DdtEz1f_rj6c_4sg/rs:fit:1500:1500/plain/gs://lumenfall-prod-assets/fderm0uzyww6xns6u35y1jzchujy@jpeg): #18 of 68 (Elo 1181)
- [Chalkboard Menu](https://assets.lumenfall.ai/YW_iu6OmqeOT3l7vinHRBuf4G_4JKJHnJqeBaKF7y4Q/rs:fit:1500:1500/plain/gs://lumenfall-prod-assets/5pjnmdb4swmz8uovc5yxt3ctm5ny@jpeg): #28 of 69 (Elo 1079)

## Example Prompt

The following prompt was used to generate an example image in our playground:

A cozy indie bookstore storefront with a large gold-leaf sign on the glass that reads "THE NOVEL NOOK" in elegant serif typography. Through the window, a capybara is calmly sitting on a plush rug by the fireplace while a customer browses shelves.

## Code Examples

### Text to Image (/v1/images/generations)

#### cURL

curl -X POST \
  https://api.lumenfall.ai/openai/v1/images/generations \
  -H "Authorization: Bearer $LUMENFALL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "wan-2.5-preview",
    "prompt": "",
    "size": "1024x1024"
  }'

# Response:
# { "created": 1234567890, "data": [{ "url": "https://...", "revised_prompt": "..." }] }

#### JavaScript

import OpenAI from 'openai';

const client = new OpenAI({
  apiKey: 'YOUR_API_KEY',
  baseURL: 'https://api.lumenfall.ai/openai/v1'
});

const response = await client.images.generate({
  model: 'wan-2.5-preview',
  prompt: '',
  size: '1024x1024'
});

// { created: 1234567890, data: [{ url: "https://...", revised_prompt: "..." }] }
console.log(response.data[0].url);

#### Python

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_API_KEY",
    base_url="https://api.lumenfall.ai/openai/v1"
)

response = client.images.generate(
    model="wan-2.5-preview",
    prompt="",
    size="1024x1024"
)

# { created: 1234567890, data: [{ url: "https://...", revised_prompt: "..." }] }
print(response.data[0].url)

### Image Edit (/v1/images/edits)

#### cURL

curl -X POST \
  https://api.lumenfall.ai/openai/v1/images/edits \
  -H "Authorization: Bearer $LUMENFALL_API_KEY" \
  -F "model=wan-2.5-preview" \
  -F "image=@source.png" \
  -F "prompt=Add a starry night sky to this image" \
  -F "size=1024x1024"

# Response:
# { "created": 1234567890, "data": [{ "url": "https://...", "revised_prompt": "..." }] }

#### JavaScript

import OpenAI from 'openai';
import fs from 'fs';

const client = new OpenAI({
  apiKey: 'YOUR_API_KEY',
  baseURL: 'https://api.lumenfall.ai/openai/v1'
});

const response = await client.images.edit({
  model: 'wan-2.5-preview',
  image: fs.createReadStream('source.png'),
  prompt: 'Add a starry night sky to this image',
  size: '1024x1024'
});

// { created: 1234567890, data: [{ url: "https://...", revised_prompt: "..." }] }
console.log(response.data[0].url);

#### Python

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_API_KEY",
    base_url="https://api.lumenfall.ai/openai/v1"
)

response = client.images.edit(
    model="wan-2.5-preview",
    image=open("source.png", "rb"),
    prompt="Add a starry night sky to this image",
    size="1024x1024"
)

# { created: 1234567890, data: [{ url: "https://...", revised_prompt: "..." }] }
print(response.data[0].url)


## About

## Overview
Wan 2.5 (Preview) is a high-performance image generation model developed by Alibaba’s Wan AI team. It is designed for both text-to-image and image-to-image workflows, focusing on high-fidelity visual output and nuanced prompt adherence. This preview release represents Alibaba's latest advancement in generative modeling, aiming to compete with leading diffusion models by balancing computational efficiency with aesthetic quality.

## Strengths
*   **Prompt Adherence:** The model demonstrates a strong ability to follow complex, multi-part descriptive prompts, accurately placing objects and maintaining specified color palettes.
*   **Image-to-Image Versatility:** Beyond generating images from scratch, it excels at taking reference images and applying stylistic or structural modifications while preserving the essence of the source material.
*   **Compositional Detail:** It is particularly effective at rendering scenes with realistic lighting, shadows, and textures, reducing the common "plastic" look sometimes found in earlier diffusion iterations.
*   **Text Rendering:** Its architecture shows improved reliability in rendering legible text within generated images compared to older generation models in the same class.

## Limitations
*   **Sensitivity to Short Prompts:** As a preview model, it often performs best with detailed descriptions; very brief or ambiguous prompts may lead to generic or unpredictable results.
*   **Anatomical Accuracy:** Like many current diffusion models, it can occasionally struggle with complex human anatomy, such as intricate hand positions or high-action poses, requiring iterative prompting to resolve.
*   **Regional Latency:** Depending on the provider infrastructure, inference times may be slightly higher than lightweight distilled models, making it less suitable for real-time applications.

## Technical Background
Wan 2.5 is part of the Wan AI suite and utilizes a diffusion-based architecture optimized for high-resolution synthesis. The model is trained on a massive dataset of high-quality image-text pairs, employing specific training techniques to enhance spatial reasoning and visual consistency. While specific architectural whitepapers for this preview release are forthcoming, it follows the transformer-based diffusion paradigm (DiT) that has become the standard for modern high-performance generative AI.

## Best For
*   **Creative Asset Generation:** Ideal for designers needing concept art, marketing visuals, or high-fidelity backgrounds with precise control.
*   **Style Transfer and Editing:** Strong for workflows where a user needs to transform an existing image into a different aesthetic or update specific elements of a composition.
*   **Prototyping:** Useful for developers building applications that require high-quality visual outputs for user-facing content.

Wan 2.5 (Preview) is available for immediate testing through **Lumenfall’s unified API and interactive playground**, allowing you to integrate it into your production environment or experiment with its capabilities alongside other leading models.

## Frequently Asked Questions

### How much does Wan 2.5 (Preview) cost?

Wan 2.5 (Preview) starts at $0.05 per image through Lumenfall. Pricing varies by provider. Lumenfall does not add any markup to provider pricing.

### How do I use Wan 2.5 (Preview) via API?

You can use Wan 2.5 (Preview) through Lumenfall's OpenAI-compatible API. Send requests to the unified endpoint with model ID "wan-2.5-preview". Code examples are available in Python, JavaScript, and cURL.

### Which providers offer Wan 2.5 (Preview)?

Wan 2.5 (Preview) is available through fal.ai on Lumenfall. Lumenfall automatically routes requests to the best available provider.

## Links

- Model Page: https://lumenfall.ai/models/alibaba/wan-2.5-preview
- About: https://lumenfall.ai/models/alibaba/wan-2.5-preview/about
- Providers, Pricing & Performance: https://lumenfall.ai/models/alibaba/wan-2.5-preview/providers
- API Reference: https://lumenfall.ai/models/alibaba/wan-2.5-preview/api
- Benchmarks: https://lumenfall.ai/models/alibaba/wan-2.5-preview/benchmarks
- Use Cases: https://lumenfall.ai/models/alibaba/wan-2.5-preview/use-cases
- Gallery: https://lumenfall.ai/models/alibaba/wan-2.5-preview/gallery
- Playground: https://lumenfall.ai/models/alibaba/wan-2.5-preview/playground
- API Documentation: https://docs.lumenfall.ai