Wan 2.5 Text-to-Image

About Wan 2.5 Text to Image

Wan 2.5 text to image produces one still per request at 45 credits. It is a text-only page: there is no upload field, so the prompt is the input, alongside the pixel size, a negative prompt, a seed and a prompt-expansion switch. The form opens on 1024×1024, and the finished image downloads as a PNG.

Sizes are eight fixed pixel pairs rather than a ratio menu. Two are square, at 512×512 and 1024×1024; three read portrait, at 768×1024, 768×1344 and 1024×1536; three read landscape, at 1024×768, 1344×768 and 1536×1024. The price does not move with the size: 45 credits per request, shown on the form and charged once whether you pick the smallest square or the tallest portrait. A wide hero and a square feed tile cost the same, so the crop should come from the layout, not from the budget.

The prompt is the main input, and the page sets no character ceiling of its own, so the limit is what you can keep readable rather than a counter. A negative prompt takes a list of things to avoid, which is where recurring blur, extra fingers or an unwanted palette belong. Prompt expansion is a switch, off by default, that lets the model add descriptive detail to what you wrote; it helps a sparse prompt and gets in the way when exact wording matters. The seed field takes -1 for a random result or a fixed integer, and a fixed seed with the same prompt and size is how one composition is iterated without losing the version that worked. Changing only the negative prompt or the expansion switch is enough to make two runs diverge. A prompt that only names a subject leaves the model to choose framing and light, so a lighting note in the prompt is a direct instruction while expansion is the model's own addition.

At 45 credits the page sits between the cheap end of the Flux 2 line, which runs from 6 to 27 credits across Dev, Pro and Flex, and Nano Banana Pro at 240. It is plain text to image: nothing is uploaded and no existing picture is edited here, so a reference-led job belongs on another page. Output appears in the preview panel as a downloadable PNG and can be opened full size before saving. Each request returns one image, so a set means several runs rather than a count field. A run the provider fails to return is refunded to the account.

How Wan 2.5 Text to Image works

1. 1

   Pick a pixel size

   Eight options from 512×512 to 1536×1024, covering square, portrait and landscape. The charge is the same 45 credits at every size, and the form starts at 1024×1024.

2. 2

   Write the prompt

   Describe the subject, style and lighting, then list what should not appear in the negative prompt. Decide whether prompt expansion stays off; leaving it off keeps the wording as written.

3. 3

   Fix the seed and generate

   Leave the seed at -1 for a random image or enter a number to repeat a composition. Confirm the 45-credit charge, run the request, and download the PNG from the preview panel.

What Wan 2.5 Text to Image is best for

• Poster and hero crops

  768×1344 and 1536×1024 give a tall or wide frame without generating square and cropping later, at the same 45 credits as any other size.

• Square feed tiles

  1024×1024 costs the same as every other size, so a square tile carries no premium over a full poster and the two can be made in one sitting.

• Iterating on one composition

  Hold the seed, prompt and size and the model has what it needs to return to the same composition with one detail changed, instead of restating the whole scene from scratch.

• Cleaning up recurring artifacts

  Name the defects you keep getting, such as blur or distorted anatomy, in the negative prompt and keep the run that stops producing them; the list carries into the next attempt.

Wan 2.5 Text to Image at a glance

Model

Alibaba Wan 2.5 text to image, run through WaveSpeed

Credit cost

45 per request, shown on the form

Sizes

512×512, 1024×1024, 768×1024, 768×1344, 1024×1536, 1024×768, 1344×768, 1536×1024

Controls

Prompt, negative prompt, seed (-1 for random), prompt expansion

Input

Text only; the page has no upload field

Output

PNG, downloaded from the preview panel

Wan 2.5 Text to Image FAQ

How much does a Wan 2.5 text to image run cost?

45 credits per request, whatever size you choose, and the figure sits on the form before you run it. A failed generation is refunded. New accounts start with 300 credits and can claim 50 more on each day they log in, both expiring after 30 days; Starter is $4.90 a month for 300 credits.

Which size should I choose?

1024×1024 or 512×512 for a square post, 768×1024 and 1024×1536 for a portrait crop, and 1344×768 or 1536×1024 for a wide one. All eight cost the same, so pick the crop the layout needs rather than the cheapest option; 768×1344 is the tallest frame on the page.

How do I get the same image twice?

Record the seed, the prompt, the size and the prompt-expansion setting. A fixed seed with the other three unchanged is what makes a rerun comparable; -1 gives a random result each time, and editing the negative prompt is enough to move the image.

Can I use a reference image?

Not on this page. It is text to image with no upload field, so a reference-image workflow belongs on a different route. Wan 2.5 image generation here starts from words only, and the negative prompt is the closest thing to steering an existing look.