wan2.7-image-pro Is Now DashScope's Recommended Image API: A Routing Decision Guide for Operators

Alibaba Cloud's DashScope has quietly updated its image generation catalog to make wan2.7-image-pro the recommended default for most image generation and editing tasks — a direct change to the model selection logic for any routing layer or AI gateway targeting DashScope endpoints.

This matters for operators because wan2.7-image-pro is not a minor version bump. It is the only model in the current DashScope catalog that combines text-to-image, image editing, bounding-box interactive editing, character-consistent multi-image generation, text rendering in images, and brand color control — all under a single model ID. For teams that previously split generation and editing across qwen-image-2.0-pro (edit) and qwen-image-max (generate), the routing configuration needs to change.

What DashScope Now Recommends

The official DashScope image model selection guide (help.aliyun.com/zh/model-studio/image-model) now leads with this:

"We recommend using wan2.7-image-pro — it integrates text rendering, brand color control, character-consistent multi-image generation, and image editing in a single model."

That is a meaningful consolidation. The model supports:

• Text-to-image generation up to 4096×4096 resolution (the highest in the catalog — all other models cap at 2048)
• Batch generation of up to 4 images per request, up to 12 in sequential mode (enable_sequential: true)
• Image editing with up to 9 reference images as input
• Bounding-box interactive editing (bbox_list parameter) to target specific regions
• A thinking_mode parameter that engages a reasoning layer before generation — relevant for complex scene layouts, multi-element brand assets, and text-heavy images
• Dual-region deployment: Beijing (dashscope.aliyuncs.com) and Singapore (dashscope-intl.aliyuncs.com), with separate API keys per region

When to Route to Which Model — A Decision Framework

The question for operators is not "which model is best overall" but "which model should route to which endpoint based on what this request needs." When asking when to use wan2.7-image-pro versus qwen-image-2.0-pro on DashScope, the official model matrix provides the answer:

Route to wan2.7-image-pro when your pipeline produces assets where text legibility, brand identity, or character consistency across multiple frames matters — product marketing images, multi-shot brand campaigns, text-overlaid social posts, or agentic workflows that generate an image sequence with a persistent subject.

Route to qwen-image-2.0-pro when you need negative prompts to exclude specific elements from output, or when you need more than 4 images per request (up to 6 variants in one call). Negative prompt support is the current gap in wan2.7-image-pro that makes qwen-image-2.0-pro the right fallback for content-safety-gated generation pipelines.

Route to z-image-turbo for high-volume, cost-sensitive generation where editing is not required — thumbnail generation, draft previews, bulk image variation at scale. At approximately 10× the generation speed and 1/5 the cost, it is the right model when quality is secondary to throughput.

Regional Routing: Beijing vs. Singapore Keys Are Not Interchangeable

DashScope's official documentation is explicit: Beijing and Singapore regions use separate API keys and request endpoints, and cross-region calls result in authentication failures or service errors.

This is a live operational concern. If your routing layer manages a single DASHSCOPE_API_KEY and conditionally changes the base_url, you will break calls originating from the wrong region. The endpoints are:

• Beijing: https://dashscope.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation
• Singapore: https://dashscope-intl.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation

Any provider configuration that stores DashScope credentials must account for this — either by mapping region → credential pair at the routing layer, or by confining image generation to a single region and accepting the latency profile.

Note that wan2.7-image-pro uses a non-OpenAI-compatible endpoint (/services/aigc/multimodal-generation/generation with a messages input format, not /v1/chat/completions). This means routing layers that forward requests directly to OpenAI-compatible endpoints must handle image generation traffic separately — the same architectural separation described in our earlier coverage of DashScope async media patterns.

Async vs. Synchronous: When to Use Which

The wan2.7-image-pro API reference shows an HTTP synchronous call as the primary pattern, described as "simple and recommended for most scenarios." For higher-resolution requests or complex multi-reference editing, generation latency is likely to exceed typical synchronous request budgets.

DashScope also supports an async job pattern (submit, get task ID, poll): for pipelines where a 4K output with thinking_mode: true and multiple reference images is expected to take 15-60 seconds, async scheduling avoids holding a connection open. For operators: if your routing layer does not already implement DashScope async job lifecycle management, the synchronous path with a generous timeout (90–120s) is the practical default while you build out the full polling flow.

What to Watch

The model selection logic here will evolve as DashScope adds more capabilities to wan2.7-image-pro. The current gap — negative prompt support — is the reason qwen-image-2.0-pro stays in any operator's routing config. If Alibaba adds negative prompts to the Wan model family (consistent with the trend in the Wan video generation models), qwen-image-2.0-pro loses its last routing-relevant advantage for text-to-image.

z-image-turbo is worth benchmarking against your actual quality requirements before deprioritizing it. The speed and cost profile makes it the right choice for any non-public-facing or draft-generation workload.

On TheRouter

For teams routing image generation to DashScope through TheRouter, provider configuration for DashScope must specify the correct regional base URL and credential pair. The async media routing guide covers how to handle job-based endpoints — the same pattern applies to high-latency wan2.7-image-pro requests where synchronous timeouts are insufficient. For multi-region DashScope setups, see provider configuration for how to map region-specific credentials in your routing policy.