Wan 2.7 Image

Overview

wan2.7-image/text-to-image is the text-to-image route in Alibaba's Wan series, callable on HiAPI.ai through a unified interface. It accepts Chinese and English prompts to generate images, offers optional 1K and 2K resolution tiers and eight aspect ratios, and supports a fixed random seed to reproduce the same result; with enhanced reasoning mode enabled, the model reasons first and then generates the image, trading longer generation time for image quality. It suits developers who need to produce visual assets in batches and have requirements on image specifications and result reproducibility, for production workflows such as poster illustrations, e-commerce scene images, and storyboard sketches.

Key Highlights

• Enhanced reasoning switch, choose quality or speed — thinking_mode is off by default; when enabled, the model reasons first and then generates the image, letting you switch flexibly based on your business requirements for image quality and wait time.
• Fixed seed, reproducible results — passing a seed reproduces the same image, making it easy to control variables and stabilize output when debugging prompts or generating images in batches.
• Aspect ratios from portrait to ultra-wide — besides common ratios such as 1:1, 16:9, and 9:16, it also offers 21:9, 8:1, and 1:8, directly fitting banners, vertical assets, and ultra-wide layouts.

Capabilities

Chinese and English text-to-image: prompts support Chinese and English, up to 5000 characters; developers can describe the image directly in Chinese or reuse English prompts, reducing the cost of cross-language rewriting.

Resolution and frame control: output offers 1K and 2K tiers paired with eight aspect ratios, so a single call yields a finished image matching the placement specifications, reducing post-hoc cropping and scaling.

Reasoning-enhanced generation: enabling thinking_mode improves image quality but increases time; choose by asset importance tier, for example off during drafting and on for final output.

Reproducible results: seed fixes randomness, so the same prompt and parameters repeatedly yield the same result, useful for regression comparison and result validation during batch production.

Use Cases

• Batch generation of campaign illustrations for marketing developers — enter the campaign theme and image description, generate at 16:9 or 1:1, and produce illustrations ready for landing pages and ad placements.
• Product scene images for e-commerce systems — use product selling-point copy as prompts to generate scene images as display assets for product detail pages and listing pages.
• Storyboard and shot sketches for content teams — write prompts shot by shot from the script and fix the seed to produce reproducible storyboard reference images for early creative review.
• Multi-placement asset adaptation — generate the same description at ratios such as 21:9, 3:4, and 1:8 to produce complete asset sets for banner, vertical, and ultra-wide placements.

Limitations and Selection Advice

This route is pure text-to-image: it does not accept reference image input and cannot provide image editing or inpainting. Output resolution is capped at 2K; for 4K images, use the Pro route in the same series. If the task depends on multi-image consistency, instruction-based editing, or click-to-edit, this route cannot cover it. Text rendering in prompts is not guaranteed, and finished images with precise layout and brand color requirements still need manual review. Generation results are heavily influenced by prompt wording, so it is recommended to test with a lower tier and a fixed seed first, then generate in batches.