Qwen Image 2.1 vs Qwen Image 3.0 vs Seedream 5.0 Flash vs Wan 2.7 Image: three prompts

Disclosure: This article was written with AI assistance and reviewed by EmpirioLabs AI.

Short answer: all four models rendered the requested sign text correctly. Seedream 5.0 Flash was the fastest, and the lowest-priced of the variants tested at 2K. Qwen Image 2.1 Pro cost less at 2K than the Pro variants of Qwen Image 3.0 and Wan 2.7 Image, and it takes up to ten reference images. All four run on EmpirioLabs through the same image endpoint. Compare the results below; we are not declaring a winner.

Test 1: sign text

Prompt: "A hand-painted wooden sign outside a seaside bakery. The sign reads FRESH SOURDOUGH DAILY in large cream letters, and a smaller line beneath it reads Open 7am to 3pm. Morning light, a gull on the roof, the sea in the background."

Test 2: following the prompt

Prompt: "Three glass jars on a wooden kitchen shelf. From left to right: a red jar filled with buttons, a blue jar filled with marbles, and a yellow jar filled with seashells. A small white cat sits to the right of the yellow jar, looking at the camera."

Test 3: low light

Prompt: "Close-up photograph of an elderly fisherman mending a net at dusk, lit only by a hanging oil lantern, weathered skin and rope texture in sharp detail, shallow depth of field, 85mm lens."

Specs and results at a glance

How we ran it

Each model received the identical prompt for each test, one shot, at its default variant, 2K resolution and a 1:1 aspect ratio, with every other setting at its default. Every image came back at 2048x2048 and is shown as returned, scaled down for the page. Prices are the live per-image rates for those settings. Times are wall-clock seconds for one synchronous request on October 18, 2026.

What to look for

Text. All four spelled both lines of the sign as written. Compare how each sets the smaller line and how the lettering sits on the wood.

Prompt following. All four kept the jars in order with the cat to the right. Qwen Image 2.1 read "a red jar" as a red lid on clear glass, while the other three colored the glass itself. Qwen Image 3.0 added labels to the jars, and in the third test a sign, that the prompts did not ask for.

Light. The third prompt asks for a scene lit only by the lantern. Compare how much light each model adds from the sky and the background.

Run the same test on EmpirioLabs

curl -X POST https://api.empiriolabs.ai/v1/images/generations \
  -H "Authorization: Bearer $EMPIRIOLABS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen-image-2-1",
    "prompt": "A hand-painted wooden sign outside a seaside bakery. The sign reads FRESH SOURDOUGH DAILY in large cream letters, and a smaller line beneath it reads Open 7am to 3pm.",
    "resolution": "2K",
    "aspect_ratio": "1:1",
    "sync": true
  }'

Swap model to qwen-image-3-0 or wan2-7-image to run the same request against the others. For seedream-5-0-flash, send quality: "2K" in place of resolution. Or try them in the playground.

Frequently asked questions

Were the images edited or retried?

No. Each image is the first one the model returned for that prompt, unedited apart from scaling. A request that returned no image was sent again.

Which model costs the least at 2K?

Of the variants tested, Seedream 5.0 Flash at $0.030 per image. Qwen Image 2.1 Turbo, not tested here, costs $0.016 per image.

Which models accept the most reference images?

Qwen Image 2.1 and Seedream 5.0 Flash accept up to 10, Wan 2.7 Image up to 9, and Qwen Image 3.0 up to 3.

Try it

Playground | All models | Pricing