Seedream v3 vs Pruna P-Image vs Wan Image Small: 5 Prompt Text to Image Test
Seedream v3 is a strong baseline for text-to-image. But fast models can surprise. This 5 prompt test compares Seedream v3, Pruna P-Image, and Wan Image Small with the same prompts.
Models in this test
- ByteDance Seedream v3 (text-to-image)
- Pruna P-Image (text-to-image, fast)
- Pruna Wan Image Small (text-to-image, small and fast)
Test setup
- 5 prompts, 1 image per model per prompt.
- Seed fixed per prompt for repeatability.
- Seedream v3 size: 1024×1024.
- P-Image aspect ratio: 1:1 (outputs came back as 1024×1024).
- Wan Image Small aspect ratio: 1:1, output format: jpg (outputs came back as 720×720).
Prompt results
Prompt 1: Studio product shot with exact text
Prompt: Photoreal studio product photo of a matte black travel mug on a concrete pedestal. Add clear legible text on the mug: WIRO API. Softbox lighting. Sharp focus. Shallow depth of field.
| Seedream v3 | Pruna P-Image | Wan Image Small |
|---|---|---|
 |
 |
 |
- All three models nailed the exact text WIRO API on this one.
- P-Image and Seedream v3 kept a clean product-photo look.
- Wan Image Small held up well, but the output resolution was lower.
Prompt 2: Street photo with Turkish neon sign text
Prompt: Documentary photo of a busy street food stall in Istanbul at night. Steam rising from a grill. Neon sign with clear Turkish text: TAZE SIMIT. Wet pavement reflections. Natural faces. 35mm film look.
| Seedream v3 | Pruna P-Image | Wan Image Small |
|---|---|---|
 |
 |
 |
- P-Image produced the cleanest readable TAZE SIMIT signage in this test.
- Seedream v3 kept the vibe, but the sign text drifted (spelling was off).
- Wan Image Small struggled with the exact lettering and produced garbled text.
Prompt 3: Minimal infographic with three exact labels
Prompt: Clean minimalist infographic poster on a white background. Three rounded rectangles in a vertical stack with crisp text. Top box says 200 OK in green. Middle box says 404 NOT FOUND in orange. Bottom box says 500 ERROR in red. Flat design. Perfect typography.
| Seedream v3 | Pruna P-Image | Wan Image Small |
|---|---|---|
 |
 |
 |
- Seedream v3 hit the clean typography target and kept all three labels readable.
- P-Image kept the right words, with minor casing drift on OK.
- Wan Image Small dropped the 500 label and rendered the last line as ERRO.
Prompt 4: Surreal glass koi above a wet road
Prompt: Surreal cinematic photo. A transparent glass koi fish swims through the air above a rain soaked asphalt road. Headlights create sharp reflections in the glass. Water droplets frozen in motion. 50mm lens.
| Seedream v3 | Pruna P-Image | Wan Image Small |
|---|---|---|
 |
 |
 |
- Seedream v3 matched the prompt closely and delivered the fish above the road look.
- P-Image shifted the idea into a glass bowl setup with multiple koi.
- Wan Image Small kept the rainy night mood, but changed the staging (fish behind glass).
Prompt 5: Astronaut barista portrait with apron and latte art
Prompt: Cinematic portrait. An astronaut barista inside a tiny cafe on Mars. Spacesuit with a simple apron. Making a latte with latte art shaped like a planet. Warm morning light through a window. Detailed face behind the visor.
| Seedream v3 | Pruna P-Image | Wan Image Small |
|---|---|---|
 |
 |
 |
- Seedream v3 included the apron detail and showed latte art clearly.
- P-Image and Wan Image Small kept the astronaut barista concept, but dropped the apron detail.
Scorecard
| Model | Text rendering (3 prompts) | Prompt adherence (5 prompts) | Typical run time in this test |
|---|---|---|---|
| Seedream v3 | Strong on clean labels, mixed on natural sign spelling | Strong | About 9 to 13 seconds per image |
| Pruna P-Image | Best street sign result here, small casing drift on OK | Good, with occasional scene restaging | About 10 to 13 seconds per image |
| Wan Image Small | Struggled with exact lettering and dropped labels | Good mood, weaker precision | About 10 to 24 seconds per image |