Alibaba quietly rolled out Wan 2.7's image generation suite, dropping four models on its Wan platform and through third-party providers. The catch: this was supposed to be the big Wan 2.7 Video launch, promised for March. Instead, April brought image-only models. No video, no open weights.
The lineup splits into two tiers. Standard models cover text-to-image and image editing at typical resolutions. Pro variants push output to 4K (4096x4096), priced higher but aimed at print and large-format work. All four models feature what Alibaba calls "thinking mode," a built-in chain-of-thought reasoning step that runs before generation. According to WaveSpeedAI's listing, this is meant to improve prompt adherence and reduce artifacts, though independent benchmarks haven't surfaced yet.
The image editing models accept up to nine reference images per request for style, subject, and background guidance. Alibaba also claims improved face rendering, HEX color code support, small text generation, and the ability to produce up to 12 consistent images from a single prompt. The models are accessible via Alibaba Cloud's Model Studio API. No local weights, no ComfyUI integration. API-only for now.
Open weights for the broader Wan 2.7 family could follow in mid-to-late Q2 2026 based on Alibaba's past release pattern, but nothing is confirmed. The video models everyone was waiting for remain unannounced.
Bottom Line
Wan 2.7 ships image-only for now with four API endpoints, thinking-mode reasoning, and 4K Pro output, but the anticipated video models and open weights are still missing.
Quick Facts
- Four models: Text-to-Image, Image-Edit, Text-to-Image-Pro (4K), Image-Edit-Pro (4K)
- Pro tier supports up to 4096x4096 resolution
- Up to 9 reference images accepted per editing request
- Up to 12 consistent images from a single prompt (company-reported)
- API-only access, no open weights released
