One model. Two superpowers. Generate stunning images from text or edit existing ones — with built-in Thinking Mode, 9-reference inputs, and 4K Pro output. No app switching: the same model handles both generation and editing with equal precision.
No app switching. Generate or edit — the same model handles both with equal precision.
Describe what you want and get a high-quality image. Thinking Mode reasons through complex prompts for better composition, accurate spatial relationships, and fewer artifacts. Supports up to 5,000 characters of prompt detail, up to 4K resolution (Pro), image set mode (up to 12 images), seed-based reproducibility, and flexible aspect ratios.
Upload up to 9 reference images and describe your changes in plain words. The model edits, restyles, or fuses images together — style transfer, element swapping, background replacement, or blending multiple references into one output, all while preserving subject and structure.
Real outputs generated with this model — every caption is the exact prompt used.
A boutique tea house at dusk — its hand-lettered sign glowing in French, Chinese, and Japanese, steam curling from the window.
An ultra-realistic studio portrait of a weathered fisherman — every pore and sun-line rendered in soft Rembrandt light.
A terracotta ceramic pour-over dripper on oatmeal linen with a single brass spoon — precise, catalog-perfect color.
A wide, golden-hour bookstore-café — spiral staircase, barista, and staggered tables in balanced cinematic depth.
An extreme macro of a dew-laden dragonfly at dawn — compound eyes and iridescent wings in crystalline detail.
A character turnaround of an original copper-goggled inventor — four perfectly consistent views, one identity.
Write your own prompt and generate high-resolution images with Wan 2.7 Image. Start free — no credit card required.
Try Wan 2.7 ImageWan 2.7 Image is Alibaba's latest unified AI model for image generation and editing, released in April 2026. It handles both creating new images from text prompts and editing existing images — in a single model, without switching tools. Key capabilities include built-in Thinking Mode for enhanced reasoning, up to 9 reference images for editing, 12-language text rendering, and 4K output in the Pro tier.
For text-to-image, you describe what you want and get a high-quality image. Thinking Mode reasons through complex prompts for better composition, accurate spatial relationships, and fewer artifacts, with support for up to 5,000 characters of prompt detail.
For image editing, you upload up to 9 reference images and describe your changes in plain words. The model edits, restyles, or fuses images together — style transfer, element swapping, background replacement, or blending multiple references into one output.
Built-in chain-of-thought reasoning that works before generating the image. When enabled, the model reasons through spatial relationships, composition balance, and multi-element prompts before committing to output — delivering better prompt adherence and fewer artifacts on complex requests. Enabled by default for text-to-image.
Upload up to 9 reference images in a single editing call. The model uses them for style guidance, subject reference, background replacement, or multi-image fusion. Describe the change in plain language and the model applies it while preserving structure and subject consistency across all inputs.
Wan 2.7 Image Pro outputs up to 4096×4096 pixels — print-ready resolution for large-format assets, commercial campaigns, and high-DPI display. The standard model reaches 2K (2048×2048), which covers most digital and social media use cases. 4K is available for text-to-image generation.
Generate up to 12 coherent, related images from a single prompt in one request. Ideal for creating a character across different scenes, product shots from multiple angles, or storyboard sequences. All images in a set share consistent style, lighting, and subject identity — no manual consistency work needed.
Text inside images renders crisply and print-ready across 12 languages, with support for up to 3,000 tokens of text content. Precise HEX color matching lets you specify exact brand colors, and palette extraction from reference photos ensures color accuracy across design systems.
Fine-grained facial control lets you dial in every detail — bone structure, eye shape, skin contours, and texture. Wan 2.7 Image excels at generating realistic faces with instruction-following precision, making it particularly strong for portrait work, character design, and brand persona creation.
Set a seed value to reproduce any generation exactly. Combined with flexible sizing options (1K, 2K, 4K, or custom dimensions), this makes Wan 2.7 Image ideal for iterative workflows — test variations, lock in the best result, and reproduce it at any scale without losing the composition.
Specify exact HEX color codes in your prompt or extract palettes directly from reference photos. The model applies these colors with precision across the generated image — critical for brand consistency in commercial work, product photography, and design systems where color accuracy is non-negotiable.
Pass multiple reference images and describe how to blend them. The model can apply the style of one image to the subject of another, swap elements between references, or fuse multiple inputs into a single coherent output. Particularly effective for adapting existing assets to new visual contexts.
| Type | Unified Image Generation & Editing |
|---|---|
| Reasoning | Built-in Thinking Mode (CoT) |
| Developer | Alibaba (Wan AI) |
| Release | April 2026 |
| License | Commercial cloud API |
| Standard | Up to 2K (2048×2048) |
|---|---|
| Pro | Up to 4K (4096×4096) |
| Size Options | 1K, 2K, 4K, custom |
| Batch Output | 1–4 (standard), 1–12 (set mode) |
| Formats | JPG, PNG, WebP |
| Prompt Length | Up to 5,000 characters |
|---|---|
| Reference Images | Up to 9 per request |
| Image Formats | JPG, PNG, BMP, WebP |
| Text in Image | Up to 3,000 tokens |
| Languages | 12 languages supported |
| Type | Chain-of-Thought reasoning |
|---|---|
| Default | Enabled for text-to-image |
| Applies to | Text-to-image only |
| Trade-off | Higher quality, longer generation |
| Best for | Complex, multi-element prompts |
| Max Images | Up to 12 per request |
|---|---|
| Consistency | Shared style, lighting, subject |
| Use Cases | Character sheets, product angles |
| Storyboards | Sequential scene generation |
| Prompt | Structured multi-image descriptions |
| Seed Range | 0–2,147,483,647 |
|---|---|
| Color Control | HEX codes, palette extraction |
| Facial Detail | Bone structure, skin, eye shape |
| Editing | Plain language instructions |
| Visual Pointing | Point-and-describe editing |
Generate brand-consistent imagery with exact HEX color matching and palette extraction from reference photos. 4K Pro output ensures assets stay sharp across all display formats — from social media to billboard-scale print.
Use image set mode to generate product shots from multiple angles in a single request. Style transfer lets you adapt existing product photos to new backgrounds or contexts without reshooting. Consistent lighting and color across all outputs.
Fine-grained facial control makes Wan 2.7 Image exceptional for portrait work. Dial in bone structure, eye shape, and skin contours with instruction precision. Create consistent character sheets across multiple poses and expressions using image set mode.
Generate slides, infographics, and data visualizations with crisp multi-language text rendering. Thinking Mode handles complex layout prompts — describe a slide structure and the model reasons through composition before generating. Up to 3,000 tokens of text content per image.
Generate up to 12 sequential scene images in one request with image set mode. Consistent characters and environments across all frames make Wan 2.7 Image ideal for pre-production storyboards, concept art series, and narrative illustration projects.
Seed-based reproducibility and flexible sizing make Wan 2.7 Image reliable for automated pipelines. Generate, test variations, lock in the best seed, and reproduce at any resolution. Batch generation (1–12 images per request) reduces API calls for high-volume workflows.
Choose the tier that fits your quality and output requirements
| Feature | Wan 2.7 Image | Wan 2.7 Image ProNEW |
|---|---|---|
| Max Resolution | 2K (2048×2048) | 4K (4096×4096) |
| Thinking Mode | Yes | Yes |
| 4K Text-to-Image | No | Yes |
| Image Editing | Up to 9 images | Up to 9 images |
| Image Set Mode | Up to 12 images | Up to 12 images |
| Best For | Everyday creation, fast iteration | Print, large-format, commercial |
Thinking Mode significantly increases generation time — around 51 seconds for a single image in Pro. For high-volume, time-sensitive workflows, disabling Thinking Mode trades some quality for speed.
When using more than 4 reference images, characters from all inputs may not blend smoothly. The model works best with 1–4 references for clean fusion; 5–9 references can produce inconsistent mixing between subjects.
Unlike earlier Wan video models that released open weights on GitHub, Wan 2.7 Image is currently cloud-only. Self-hosted deployment is not yet available, which limits use cases requiring on-premise or air-gapped environments.
4K resolution is only available for text-to-image generation — not for image editing mode. When editing with reference images, the maximum output is 2K even on the Pro tier. Plan your workflow accordingly for large-format editing tasks.
Wan 2.7 Image applies content moderation filters. The model excels at realistic photo generation but enforces content policies that restrict certain categories of imagery. This is standard for cloud-hosted commercial models.
4K output images can reach 25MB per file. For workflows generating large batches at maximum resolution, storage and transfer costs add up quickly. Consider using 2K for digital-only outputs and reserving 4K for final print-ready assets.
Experience Alibaba's unified AI image model. Thinking Mode reasoning, 9-reference editing, 12-language text, and 4K Pro output — all in one place on SharkFoto.
Try Wan 2.7 Image now