---
title: "Seedream 4.0 Blind Test Results: Scores by Use Case"
canonical_url: "https://workroom.everypixel.com/models/seedream-4-0"
score: 56.9
score_ci: [44, 70]
rank: 4
scale: "Anchor comparison of generative image models, v3"
tested: 2026-09-11
revised: 2026-09-14
---

# Seedream 4.0 Blind Test Results: Scores by Use Case

**Score:** 56.9 (95% interval 44–70). A real photograph is 100.
**Rank:** #4 on the scale
**Author:** Dmitry Shironosov (Head of AI Research, Everypixel Workroom)
**Tested:** 2026-09-11 · revised 2026-09-14
**Canonical URL:** https://workroom.everypixel.com/models/seedream-4-0

## Independence

Workroom is an AI router: we give access to third-party models, including Seedream 4.0, and we are not affiliated with ByteDance. We have no stake in any single vendor winning; the router is only useful if it sends each brief to the right model. The judgement-level data behind this card is published by Everypixel research under CC BY 4.0 and linked below, so every number here can be checked independently.

## TL;DR

Seedream 4.0 scores 57 on our photo-anchored scale (95% interval 44–70), where a real photograph is 100. In 864 blind judgements on 48 commercial scenes it showed no measurable difference from Qwen Image 2.0: Seedream was preferred in 48.2% of judgements, ties split. GPT Image 2 was clearly ahead, preferred in 68.6%. By category, Seedream leaned ahead on social marketing and fashion, and behind on product shots, typography and street scenes.

## Where it stands

| Rank | Model | Score | 95% interval |
|---|---|---|---|
| #1 | GPT Image 2 | 104 | — |
| — | Real photograph | 100 | — |
| #2 | Qwen Image 2.0 Pro | 80 | 68.6–93.8 |
| #3 | Qwen Image 2.0 | 65 | — |
| #4 | Seedream 4.0 (this card) | 57 | 44–70 |
| #5 | FLUX.2 Klein | 24 | — |

Axis: believability. A real photograph is 100. A score above 100 is the scale's technical ceiling and does not mean parity.

### By use case

| Category | Score | Rank |
|---|---|---|
| Interior | 83 | #4 |
| Fashion and beauty | 80 | #2 |
| Social marketing | 79 | #3 |
| Product and e-commerce | 67 | #4 |
| Food and beverage | 57 | #3 |
| Street and architecture | 53 | #4 |
| People and lifestyle | 16 | #3 |
| Layout and typography | −14 | #4 |

*Category positions come from 6 scenes each — a tendency, not a measurement. A real photograph is 100 on every ruler.*

## When to use as default

Route social-marketing and fashion briefs to Seedream 4.0: those were the categories where it leaned ahead of Qwen Image 2.0 in our blind test. With six scenes per category this is a tendency rather than a guarantee, so for an important brief keep a second take from Qwen Image 2.0.

## When to use as finisher

Not a finisher. When the final asset has to be the most believable image, finish in GPT Image 2: it was preferred over Seedream 4.0 in 68.6% of judgements, with a believability gap of +0.78 [+0.55, +1.03]. Seedream works as a second mid-tier take next to Qwen Image 2.0 — the two are statistically level.

## When to look elsewhere

Look elsewhere for product and e-commerce shots, layouts with typography, and street or architecture scenes: Seedream 4.0 fell behind Qwen Image 2.0 on all three. It also does not accept a seed, so a result cannot be regenerated exactly.

## Direct blind tests

| Opponent | Result | Gap (Bradley–Terry, 95% CI) | Preferred in | Judgements |
|---|---|---|---|---|
| Qwen Image 2.0 | A coin flip: 48% for Seedream 4.0, no measurable difference. | −0.07 [−0.28, +0.14] | 48.2% (ties split, all questions) | 432 |
| GPT Image 2 | GPT Image 2 was preferred over Seedream 4.0 in 69% of blind comparisons. | −0.78 [−1.03, −0.55] | 31.4% (ties split, all questions) | 432 |

## Price

| Provider | Price | Checked |
|---|---|---|
| [ByteDance (BytePlus ModelArk)](https://docs.byteplus.com/en/docs/ModelArk/1824718) | $0.03 / img | 2026-09-25 |
| [fal.ai](https://fal.ai/models/fal-ai/bytedance/seedream/v4/text-to-image) | $0.03 / img | 2026-09-25 |
| [Together AI](https://www.together.ai/models/bytedance-seedream-4-0) | $0.03 / img | 2026-09-25 |
| [Runware](https://runware.ai/docs/models/bytedance-seedream-4-0) | $0.03 / img | 2026-09-25 |

Average: $0.03 / img across 4 providers

Prices are for one 1024×1024 image at standard settings, checked 25 September 2026; the price at the top of the page is the average of the providers listed. Workroom pricing is on the pricing page.

## License

Seedream 4.0 is a proprietary ByteDance model served through the BytePlus ModelArk API; its weights are not published. Terms for generated images are set by the service you generate them in.

## FAQ

### Is Seedream 4.0 better than Qwen Image 2.0?

No measurable difference. On believability the Bradley–Terry gap is −0.07 with a 95% interval from −0.28 to +0.14, which includes zero. Across all three questions Seedream 4.0 was preferred in 48.2% of judgements with ties split, close to a coin flip. The two models differ by category rather than overall.

### How far behind GPT Image 2 is Seedream 4.0?

Clearly behind. The believability gap is −0.78 with a 95% interval from −1.03 to −0.55, which does not include zero, and GPT Image 2 was preferred in 68.6% of judgements with ties split. On the photo-anchored scale that places Seedream 4.0 at 57 and GPT Image 2 at 104.

### Why does GPT Image 2 score above 100 on this scale?

A real photograph is fixed at 100 by definition. GPT Image 2 lands at 104.4 because its believability strength sits slightly above the photographs', but that difference is not statistically separated. A score above 100 is the scale's technical ceiling, not a claim that the images match photography.

### What does the 44–70 range around Seedream 4.0's score mean?

It is the 95% interval of Seedream 4.0's score, with 56.9 as the point estimate. It is wide because 48 scenes still leave room for scene-to-scene variation, and it overlaps Qwen Image 2.0 at 64.5, which matches the head-to-head result of no measurable difference between the two.

### How was Seedream 4.0 tested?

We generated three independent images for each of the 48 commercial scenes the ladder was built on and paired them blind with Qwen Image 2 and GPT Image 2: 288 pairs and 864 judgements on prompt adherence, believability and aesthetics. The score uses believability only, against rungs frozen from the baseline study.

### Is Workroom affiliated with ByteDance?

No. Workroom is an AI router that gives access to third-party models, including Seedream 4.0, and is not affiliated with ByteDance. The judgement-level data behind this card is published by Everypixel research under CC BY 4.0, so every number on this page can be checked against the source dataset.

## Methodology

Between 6 and 11 September 2026 we compared Seedream 4.0 blind against two fixed rungs of our quality ladder, Qwen Image 2 and GPT Image 2, on the same 48 commercial scenes the ladder was built on — three independent generations per scene, 288 pairs, 864 judgements. Every pair was judged three times, without any indication of which image came from which model, on three questions: which image follows the prompt better, which is more believable as a real photograph, and which is more attractive.

- Scenes: 48; generations per scene: 3
- Compared with: Qwen Image 2.0, GPT Image 2
- Pairs: 288; judgements: 864; each pair judged 3 times
- Collected: 2026-09-06 … 2026-09-11
- Questions: “Which image follows the prompt better?” “Which is more believable as a real photograph?” “Which is more attractive?”
- Scale: 0.020835 strength per point; A real photograph is 100.
- Category positions are fitted from the same judgements, 6 scenes per category, on the same scale.
- The scale reproduced: frozen gap 0.831 falls inside this session's interval [0.45, 1.02]

### Caveats

- The model ignores the requested output size: Seedream returned 2352 px on the long edge where 2048 was asked. Output was downscaled to match the ladder; it was never upscaled.
- No seed control: Seedream does not accept a seed, so the three generations per scene are independent but not reproducible. This measures average ability rather than a best pick.
- Prompt adherence is biased against photography, which is why the absolute score is built on believability alone.
- Tie share is 28.4%, higher than the 25.4% of the original ladder study. This session pairs two models that are genuinely indistinguishable, where “equal” is an honest answer.



## Links

- How the scale was built: https://research.everypixel.com/anchor-ladder/
- Votes (CSV, 102 KB, CC BY 4.0): https://workroom.everypixel.com/datasets/seedream-4-0-ladder/votes.csv
- Results (JSON, 4 KB, CC BY 4.0): https://workroom.everypixel.com/datasets/seedream-4-0-ladder/results.json
- Scenes and prompts (CSV, 44 KB, CC BY 4.0): https://workroom.everypixel.com/datasets/seedream-4-0-ladder/scenes.csv
- README (MD, 2 KB, CC BY 4.0): https://workroom.everypixel.com/datasets/seedream-4-0-ladder/README.md
- Images (ZIP, 186 MB, All rights reserved — verification only): https://wr.everypixel.com/ghost-research/geo-benchmarks/datasets/seedream-4-0-evaluation-images.zip

## Citation

Everypixel research (2026). Seedream 4.0 on the anchor ladder v3 [Data set]. Ratings collected 6–11 September 2026.

Data license: CC BY 4.0. Images: All rights reserved; provided for verification only.

---
HTML version: https://workroom.everypixel.com/models/seedream-4-0
Last modified: 2026-09-14