Qwen 2 vs Wan 2.7 Pro
A recorded operational comparison from 2026-10-01. One prompt, one completed generation per model; no visual-quality scoring.
| Model | Completion time | Generation charge | Returned dimensions |
|---|---|---|---|
| Qwen 2 | 9.7 s | $0.030 | 2048 × 2048 |
| Wan 2.7 Pro | 14.0 s | $0.060 | 1024 × 1024 |
What this run shows
Qwen 2 completed sooner and had a lower recorded generation charge in this run. It also returned a larger image. These measurements say nothing about which output better followed the prompt or had better visual quality.
Do not use these single samples to predict your next request or rank overall model quality. Queue load, model changes, and settings affect results. No average, percentile, or reliability rate is inferred.
Shared prompt
A photorealistic editorial photograph of a small red ceramic teapot beside exactly two yellow lemons on a pale oak table, a folded blue linen napkin on the right, soft morning window light from the left, realistic shadows and fine surface textures, a simple cream wall, eye-level composition, no people, no logos, no text.
Recorded settings
Qwen 2
- seed
- 0
- image_size
- 1:1
- nsfw_checker
- true
- output_format
- png
Wan 2.7 Pro
- n
- 1
- seed
- 0
- watermark
- false
- resolution
- 1K
- aspect_ratio
- 1:1
- thinking_mode
- false
- enable_sequential
- false
Method and limitations
Source: Artifio’s production image-model smoke test, recorded through the normal generation flow. Completion time is the server-reported interval from generation creation to completion, including Artifio processing. It is not provider-only inference time or the time until the browser displayed the result. No reference images were supplied. Explicit prompt enhancement was disabled where exposed.
The requested square composition produced different pixel dimensions across some models. Quality controls differ, and the runs were not simultaneous. Each listed charge is the recorded generation debit, not an all-in account funding cost or a current price quote. These completed requests had no recorded generation refunds.
Customer media and account identifiers are not published. This numeric report cannot support a visual-quality comparison. A broader evaluation would require repeated runs, multiple task prompts, and separately reviewed public example outputs.