WAN V2.6 TEXT TO IMAGE

Wan v2.6 Text to Image is a text-to-image generation model available on fal.ai, designed to convert descriptive prompts into high-quality images. The model accommodates both text and optional reference images as input, enabling users to guide style and content more precisely. A key feature of Wan v2.6 is its support for prompts in both English and Chinese, making it accessible to a broader user base.

At its core, Wan v2.6 is engineered for versatility in image creation based on detailed user instructions. Users can enter a text prompt up to 2000 characters, describing the desired scene, object, or concept. The model can refine output further when provided with a reference image (via URL), which serves as a visual guide—useful for maintaining consistent style or incorporating specific elements. The reference image should be in JPEG, JPG, PNG (without alpha), BMP, or WEBP format, with each dimension between 384 and 5000 pixels and a maximum size of 10MB. Only one reference image can be supplied per request.

Flexible image sizing is a notable technical capability. Users can specify exact dimensions (height and width, from 1 pixel up to 14,142 pixels) or select from common aspect ratio presets such as 'square_hd', 'square', 'portrait_4_3', 'portrait_16_9', 'landscape_4_3', or 'landscape_16_9'. If no size is specified, the output will match the provided input image size or default to a maximum of 1280x1280 pixels.

To enhance creative control, Wan v2.6 supports a 'negative prompt' field, up to 500 characters, which lets users describe qualities or content they wish to avoid in the generated image (for example: "low resolution, error, worst quality, low quality, deformed"). The model also allows users to set a seed (integer between 0 and 2,147,483,647) for reproducibility, so the same input can yield identical results in the future.

For users requiring multiple variations, the 'max_images' parameter can be set to produce between 1 and 5 images per request, although the actual number may vary depending on model inference.

Safety and responsible use are considered, with an 'enable_safety_checker' parameter (default: true) for content moderation applied to both input and output.

The output consists of generated images in PNG format and, in certain mixed modes, can also include generated text (if enabled). The typical use case is generating new images from textual descriptions, with optional guidance from a reference image. The product supports commercial use and can be accessed interactively via a playground, or programmatically through a documented API schema.

Wan v2.6's configuration options and multilingual support position it well for a variety of creative, design, or commercial image generation tasks requiring control over content, style, output dimensions, and quality. No explicit performance, quality metrics, or unique technical limitations are documented beyond the parameter constraints and safety features outlined above.