Wan 2.5 Image
What is Wan 2.5 Image?
Wan 2.5 Image is the first Wan image model with reference support. The model auto-selects between text-to-image and editing based on whether image_urls is present. Up to three reference images, seven aspect ratios, prompt extend on by default, and seed control. Prompts are capped at 2000 characters and negative prompts at 500. It is the right step up from Wan 2.2 Image when you need to steer with existing assets without jumping to Wan 2.6 or 2.7's larger reference caps.
Key features of Wan 2.5 Image
Four features cover what changes versus Wan 2.2 Image.
Up to 3 reference images
Pass `image_urls` to switch from text-to-image into editing or image-guided mode automatically. Useful for catalog updates, style alignment, and mild restyles.
Seven aspect ratios
1:1, 2:3, 3:2, 3:4, 4:3, 9:16, and 16:9. Wider range than Wan 2.2 Image, with both portrait and landscape variants of 2:3 and 3:4.
Prompt extend on by default
The model rewrites short prompts for clarity before rendering. Turn it off when you have a tightly engineered prompt and want it preserved exactly.
2000-character prompt cap
Four times the 500-character cap on Wan 2.2 Image. Useful for detailed multi-element scene prompts and brand briefs that need every constraint stated.
Best for
Catalog updates
Adjust existing shots with one to three references and a clear instruction prompt.
Localised creatives
Tweak the same scene for different markets with prompt extend and negative prompts for brand fit.
Storyboards
Lock aspect ratio across frames so the layout stays consistent in a series.
Lightweight edits
Smaller reference set than Wan 2.6 or 2.7 Image, useful when three references are enough.
Style alignment from stills
Match brand colour and lighting from one or two reference shots.
Reproducible runs
Seed control returns the same image from the same prompt and references.
API examples
Call Wan 2.5 Image from any language by POSTing to /v1/tasks. Full parameter docs live at docs.unifically.com/models/image/alibaba/wan-2.5-image.
curl -X POST https://api.unifically.com/v1/tasks \
-H "Content-Type: application/json" \
-H "Authorization: Bearer YOUR_API_KEY" \
-d '{
"model": "alibaba/wan-2.5-image",
"input": {
"prompt": "A flower shop with a beautiful wooden door"
}
}'
Successful submission returns a task_id. Poll GET /v1/tasks/<task_id> or set a callback_url on the request to receive the finished result.
FAQs
People also ask
Wan 2.5 Image is the first Wan image model to accept reference images. It supports text-to-image, image editing, and image-guided generation with 1–3 reference images, seven aspect ratios, prompt extend by default, and a seed. Prompts are capped at 2000 characters.
Zero to three. Omit image_urls for plain text-to-image; pass 1–3 image URLs for editing or image-guided runs.
Wan 2.6 Image accepts up to four references. Wan 2.7 Image and Image Pro accept up to nine, add thinking mode for harder text-only prompts, and offer higher resolution tiers (2K standard, 4K Pro). Stay on Wan 2.5 Image when three references and this control set are enough.
Prompt extend rewrites your prompt for clarity before generation. It is on by default. Turn it off when you have a tightly engineered prompt and do not want it altered.
Seven. 1:1, 2:3, 3:2, 3:4, 4:3, 9:16, and 16:9. Default is 1:1.
Yes. Pass an integer seed alongside the prompt and any references. Re-sending the same inputs and seed returns the same image.
