WAN 2.7 Image

WAN 2.7 Image is Alibaba's Wan 2.7 image model for text-to-image, image editing, and multi-reference fusion. It supports resolutions up to about 2K (for example 2048×2048), flexible aspect ratios via preset sizes or custom dimensions, and optional coherent image-set generation for the same character or product across multiple shots. A thinking mode is available for stronger reasoning on text-to-image requests (as described on the model provider page).

On VidMachine, WAN 2.7 Image is available as an image option in your project's image model priority. We run it through Replicate as documented at replicate.com/wan-video/wan-2.7-image. Each generated image uses a fixed credit cost (see Pricing). You can use it for start frames, scene edits, thumbnails, and AI Images workflows alongside other models in your fallback chain.

For higher quality output with 4K support and a dedicated “pro” variant, Alibaba also documents WAN 2.7 Image Pro on Replicate; on VidMachine the integrated model is the standard WAN 2.7 Image variant.

Describe the scene in natural language and get a high-quality image. Prompts can be long (the upstream model supports up to thousands of characters), so you can specify composition, lighting, style, and subject detail in one go. Size can be set with presets such as 1K or 2K, or with explicit width-by-height strings for exact aspect ratios—useful for vertical shorts, landscape thumbnails, or square assets.

You can supply reference images together with a text instruction to edit, restyle, or fuse content. The model accepts multiple inputs for style transfer, element swaps, and blending several references into one output. This fits workflows where you already have a product shot, a character reference, or a mood board and want a new composite or variant without starting from a blank canvas.

Image-set mode is designed for coherent batches from one prompt—for example the same character in different seasons, product angles, or storyboard beats. The provider documents generating multiple related images in one request. On VidMachine we currently call the model for a single output per generation step, but you can still rely on the model's strength for consistency when you use it repeatedly with clear prompts and references.

Thinking mode is aimed at improved quality for text-to-image by doing more internal reasoning before rendering. A seed can be used when the API supports it for reproducible runs. Together, these help when you need to iterate on small prompt tweaks or lock down a look for a series of images.