{
 "schema_version": "1.0",
 "page_kind": "article",
 "pack_kind": "generate",
 "slug": "qwen-image-turbo-in-the-coves-style",
 "title": "Qwen-Image-2.1-Turbo, judged blind in a game's own painted style",
 "dek": "One person, picking blind, chose Qwen-Image-2.1-Turbo, a new 8-step text-to-image model, for none of six real subjects from our game in its own oil-painted sketch style, against FLUX.1 [dev] and FLUX.2 [klein] 4B: by the rules we wrote down first, not a win.",
 "published": "2026-10-18",
 "series": [
  "bench"
 ],
 "licence": "CC BY 4.0",
 "status": "pending-judge",
 "pack_note": "no judge is seated, so every chip is held",
 "notice": "items with published:false are held and are not this page's published words",
 "source": {
  "url": "https://research.strata2signal.com/qwen-image-turbo-in-the-coves-style/",
  "md_url": "https://research.strata2signal.com/qwen-image-turbo-in-the-coves-style/index.md",
  "md_sha": "5b369d4fa5f910b3d4e1ba91068dc7c6f4653e39646f60dc953cb6472cb78c49",
  "html_sha": "0ef89e474a476b6e31b317d0e59339ecbeaaadf5aece12bbe21b995c71acce24",
  "anchors_sha": "5c81288753f23c6e2016c4715a82d6ae898061186b678b4edd33c0b1076a50b1"
 },
 "short": {
  "paragraph": "Qwen-Image-2.1-Turbo painted six real subjects from our game RealKeep in the game's own sketch style, stylized oil-painted realism in the manner of the video game Dishonored, beside FLUX.1 [dev] at 32 steps and FLUX.2 [klein] 4B at 8. One person picked the best picture of each from unlabelled sheets. It is not a win. The Turbo, at either of its two sizes, was picked for 0 of the 6 subjects; FLUX.1 [dev], one of the models the game paints with today, for 4; klein 4B for 2. The Turbo did show every item on each subject's check list on 5 of the 6, as klein 4B did (FLUX.1 [dev]: 4), but a win needed both. One person's blind eye on six subjects in one style is not a ranking: it says only that the Turbo has not earned a bigger test. klein 4B was also the fastest (0.98 seconds a picture; the Turbo 1.41 at the same size, on a different clock); speed decided nothing. The Turbo's licence allows research and evaluation only, so it was never going into the game.",
  "from_dek": false,
  "counts": {
   "words": 2844,
   "minutes": 13,
   "tables": 4,
   "kit": false
  },
  "bullets": [
   {
    "text": "the hottest of the eight logged readings for the card was 62 °C",
    "figure": "62 °C",
    "cite": "how-it-ran",
    "quote": "No registered stop fired: every column made 18 of 18, and the hottest of the card's eight logged readings was 62 °C (the stop was 83).",
    "span": [
     10526,
     10532
    ]
   }
  ]
 },
 "sections": [
  {
   "id": "what-this-page-asks",
   "heading": "What this page asks",
   "level": 2,
   "span": [
    2189,
    3834
   ],
   "chunks": [
    [
     2189,
     3834
    ]
   ],
   "chars": 1645,
   "digest": "the article asks if Qwen-Image-2.1-Turbo can paint the cove's style on its real subjects at least as well as FLUX.1 [dev] without losing more subjects. it notes the model weighs 32.44 GB and its licence allows research and evaluation only.",
   "digest_skipped": null
  },
  {
   "id": "the-three-models",
   "heading": "The three models",
   "level": 2,
   "span": [
    3834,
    4843
   ],
   "chunks": [
    [
     3834,
     4843
    ]
   ],
   "chars": 1009,
   "digest": "three models are compared: Qwen-Image-2.1-Turbo, FLUX.1 [dev], and FLUX.2 [klein] 4B. the Turbo ran in Diffusers 0.41.0, while the FLUX models ran in ComfyUI v0.21.1.",
   "digest_skipped": null
  },
  {
   "id": "the-style-and-the-six-subjects",
   "heading": "The style, and the six subjects",
   "level": 2,
   "span": [
    4843,
    7034
   ],
   "chunks": [
    [
     4843,
     7034
    ]
   ],
   "chars": 2191,
   "digest": "the style is stylized oil-painted realism, loose quick brushwork, in the style of Dishonored game art. six subjects, including two foes, two scenes, and two portraits, are tested using verbatim subject clauses and specific requirements for what must show.",
   "digest_skipped": null
  },
  {
   "id": "what-counts-as-a-win",
   "heading": "What counts as a win, written down first",
   "level": 2,
   "span": [
    7034,
    8751
   ],
   "chunks": [
    [
     7034,
     8751
    ]
   ],
   "chars": 1717,
   "digest": "a win requires passing G0, G2, and G3. G2 checks if a model holds a subject, and G3 checks if the Turbo is picked for at least 4 of the 6 subjects. G4 speed is a tie-breaker.",
   "digest_skipped": null
  },
  {
   "id": "how-it-ran",
   "heading": "How it ran",
   "level": 2,
   "span": [
    8751,
    11177
   ],
   "chunks": [
    [
     8751,
     11177
    ]
   ],
   "chars": 2426,
   "digest": null,
   "digest_skipped": null
  },
  {
   "id": "what-the-judge-picked",
   "heading": "What the judge picked",
   "level": 2,
   "span": [
    11177,
    12741
   ],
   "chunks": [
    [
     11177,
     12741
    ]
   ],
   "chars": 1564,
   "digest": "the Turbo was not picked for any of the six subjects, so G3 fails. G2 passes as both Turbo columns held 5 of 6 subjects. the bench was COMPLETE.",
   "digest_skipped": null
  },
  {
   "id": "the-pictures",
   "heading": "The pictures",
   "level": 2,
   "span": [
    12741,
    14107
   ],
   "chunks": [
    [
     12741,
     14107
    ]
   ],
   "chars": 1366,
   "digest": "two picks from klein 4B are shown. the Salt-Father pick included a scrawl like a signature, despite the style clause. the Wrenna Sault pick is a close-up portrait.",
   "digest_skipped": null
  },
  {
   "id": "seconds-per-picture",
   "heading": "Seconds per picture, by each engine's own clock",
   "level": 2,
   "span": [
    14107,
    15106
   ],
   "chunks": [
    [
     14107,
     15106
    ]
   ],
   "chars": 999,
   "digest": "the median seconds per picture for the Turbo at 768 × 768 was 1.41. klein 4B was the fastest. the Turbo's median step took 0.310 s at 1024².",
   "digest_skipped": null
  },
  {
   "id": "what-this-means",
   "heading": "What this result does and does not mean",
   "level": 2,
   "span": [
    15106,
    16019
   ],
   "chunks": [
    [
     15106,
     16019
    ]
   ],
   "chars": 913,
   "digest": "the result means the Turbo did not earn a bigger test and this line of testing closes. it does not mean the Turbo would fail at other styles or at home.",
   "digest_skipped": null
  },
  {
   "id": "why-this-page-is-an-evaluation",
   "heading": "Why this page is an evaluation",
   "level": 2,
   "span": [
    16019,
    17208
   ],
   "chunks": [
    [
     16019,
     17208
    ]
   ],
   "chars": 1189,
   "digest": "the evaluation fits the Qwen Research License Agreement, which allows use for research or evaluation purposes only. the team downloaded the model once and removed the copy after the run.",
   "digest_skipped": null
  },
  {
   "id": "how-to-check-our-work",
   "heading": "How to check our work",
   "level": 2,
   "span": [
    17208,
    19243
   ],
   "chunks": [
    [
     17208,
     19243
    ]
   ],
   "chars": 2035,
   "digest": null,
   "digest_skipped": "credits"
  },
  {
   "id": "more-on-this-site",
   "heading": "More on this site",
   "level": 2,
   "span": [
    19243,
    19720
   ],
   "chunks": [
    [
     19243,
     19720
    ]
   ],
   "chars": 477,
   "digest": "the site provides links to information about who makes FLUX, the same eighteen pictures timed on different cards, and a licence ledger.",
   "digest_skipped": null
  },
  {
   "id": "who-ran-this-and-thanks",
   "heading": "Who ran this, and thanks",
   "level": 2,
   "span": [
    19720,
    20572
   ],
   "chunks": [
    [
     19720,
     20572
    ]
   ],
   "chars": 852,
   "digest": null,
   "digest_skipped": "credits"
  }
 ],
 "chips": [
  {
   "id": "c-b9bc215c",
   "q": "what is the purpose of the qwen-image-2.1-turbo model?",
   "a": "The model turns a sentence into a picture and acts as an accelerated checkpoint of Qwen-Image-2.1 for text-to-image generation and image editing using 8 denoising steps.",
   "cites": [
    "what-this-page-asks"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-042299b0",
   "q": "which models were compared in this evaluation?",
   "a": "The comparison included Qwen-Image-2.1-Turbo, FLUX.1 [dev], and FLUX.2 [klein] 4B.",
   "cites": [
    "the-three-models"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-e55006c4",
   "q": "how was the visual style defined for the subjects?",
   "a": "The style is described as stylized oil-painted realism, loose quick brushwork, in the style of Dishonored game art, unsigned, no text, no letters, no signature, no watermark.",
   "cites": [
    "the-style-and-the-six-subjects"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-8fcf19f4",
   "q": "what criteria determined a successful result?",
   "a": "A win required passing G0 (validity), G2 (showing the subject), and G3 (reading best as the specified style).",
   "cites": [
    "what-counts-as-a-win"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-61f1d24d",
   "q": "how were the different engines and clocks managed?",
   "a": "The run used two different engines and clocks, with the median over seeds 202 and 303 used for the fairest timing measurement.",
   "cites": [
    "how-it-ran"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-dad83f20",
   "q": "how did the turbo model perform in the blind test?",
   "a": "The Turbo was picked for none of the six subjects at either 768 × 768 or 1024 × 1024 resolution, failing the G3 requirement.",
   "cites": [
    "what-the-judge-picked"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-d337af2c",
   "q": "what is the median time per picture for the turbo model?",
   "a": "At 768 × 768, the median time was 1.41 seconds, and at 1024 × 1024, it was 2.68 seconds.",
   "cites": [
    "seconds-per-picture"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-54896c5c",
   "q": "what are the limitations of these findings?",
   "a": "The results do not apply to other styles, the full Qwen-Image-2.1 model, different schedules, or different hardware setups.",
   "cites": [
    "what-this-means"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  }
 ],
 "related": [
  {
   "slug": "the-new-kid",
   "why": "shares ground with § What counts as a win, written down first · § The classifier seat — five gates, all pass, faster"
  },
  {
   "slug": "two-used-3080s-priced",
   "why": "shares ground with § What this page asks · § The question"
  },
  {
   "slug": "one-3090-against-one-3090-ti",
   "why": "shares ground with § Seconds per picture, by each engine's own clock · § Renders"
  }
 ],
 "thanks": "## Who ran this, and thanks {#who-ran-this-and-thanks}\n\nThanks to Qwen, to Black Forest Labs, and to the people whose public reports on the Qwen-Image-2.1 pages told us what to look for. The Turbo ran in **Diffusers** 0.41.0 and **Transformers** 5.19.0 (Apache 2.0) on **PyTorch** 2.13.0 (BSD-style); the FLUX columns in **ComfyUI** v0.21.1 (GPL-3.0); all under **CUDA** (NVIDIA, proprietary), the one closed piece. Arkane Studios' Dishonored is named only for its style. A small human team asked for this test, set its rules first, made the blind picks and signed the numbers; a fleet of AI agents read the sources, built and ran the harness, and drafted this page under that team's rulings.\n\nCorrections and later measurements will be added below, each dated (UTC), with a window at both ends where one applies, saying in plain words what it counts.",
 "kit": null,
 "seat_class": {
  "writer": "gemma-class",
  "judge": null,
  "writer_runtime": "vllm"
 },
 "bench": {
  "bullets_written": 3,
  "bullets_kept": 1,
  "digests_written": 11,
  "digests_kept": 10,
  "chips_written": 8,
  "chips_kept": 8,
  "chips_grounded": 0,
  "chips_published": 0,
  "dropped_by": {
   "new_noun": 0,
   "figure": 0,
   "length": 0,
   "cite": 0,
   "judge": 0,
   "quote": 0,
   "directive": 0,
   "redaction": 1,
   "profanity": 0,
   "identifier": 0,
   "bare_figure": 1,
   "house_voice": 0,
   "editorial": 0,
   "hand-edit": 1
  },
  "new_noun_tokens_checked": [
   "0.310",
   "0.41.0",
   "1.41",
   "1024",
   "2.68",
   "202",
   "303",
   "32.44",
   "4",
   "4B",
   "5",
   "6",
   "6000",
   "62",
   "768",
   "8",
   "96",
   "Agreement",
   "C",
   "COMPLETE",
   "ComfyUI",
   "Diffusers",
   "Dishonored",
   "Edition",
   "FLUX",
   "FLUX.1",
   "FLUX.2",
   "G0",
   "G2",
   "G3",
   "G4",
   "GB",
   "License",
   "NVIDIA",
   "PRO",
   "Qwen",
   "Qwen-Image-2.1",
   "Qwen-Image-2.1-Turbo",
   "RTX",
   "Research",
   "Salt-Father",
   "Sault",
   "Turbo",
   "Turbo's",
   "Workstation",
   "Wrenna",
   "qwen-image-2.1-turbo",
   "v0.21.1"
  ],
  "new_noun_tokens_withheld": 1,
  "figure_definition": "v3",
  "figure_boundary": "guarded",
  "drops": [
   {
    "kind": "digest",
    "reason": "redaction",
    "item": null,
    "item_chars": 206,
    "withheld": "a string carrying a fenced literal",
    "detail": ""
   },
   {
    "kind": "bullet",
    "reason": "bare_figure",
    "item": "the Turbo was trained to copy a slower one's results in 8",
    "detail": "the cited section writes this figure as '8 denoising steps' and the bullet prints '8' alone"
   },
   {
    "kind": "bullet",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 45,
    "text_sha": "36e3e58e9cd1",
    "figure": "0.310",
    "withheld": "one bullet removed from this pack by hand",
    "removed_utc": "2026-10-17",
    "why": "the hand-read against the release source, 2026-10-17 (UTC): the figure is true (the Turbo's median step at 1024 x 1024), but the line's word order breaks the sentence, so it does not read as the page's own words",
    "detail": "removed from this pack by hand on 2026-10-17 (UTC), after the writer ran and after the gates had counted: the hand-read against the release source, 2026-10-17 (UTC): the figure is true (the Turbo's median step at 1024 x 1024), but the line's word order breaks the sentence, so it does not read as the page's own words. Not a gate drop, and recorded here so the ledger adds up."
   }
  ]
 },
 "generated_utc": "2026-10-17T02:02:52Z",
 "meta_refreshed_utc": "2026-10-17T03:25:30Z"
}