Datasets:

kimi000
/

zimage-anyflow2-vs-base-4step-showcase

License:

Dataset Viewer

The dataset viewer is not available for this subset.

Cannot get the split names for the config 'default' of the dataset.

Exception:    SplitsNotFoundError
Message:      The split names could not be parsed from the dataset config.
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
                  for split_generator in builder._split_generators(
                                         ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/folder_based_builder/folder_based_builder.py", line 251, in _split_generators
                  raise ValueError(
                      "`file_name`, `*_file_name`, `file_names` or `*_file_names` must be present as dictionary key in metadata files"
                  )
              ValueError: `file_name`, `*_file_name`, `file_names` or `*_file_names` must be present as dictionary key in metadata files
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 68, in compute_split_names_from_streaming_response
                  for split in get_dataset_split_names(
                               ~~~~~~~~~~~~~~~~~~~~~~~^
                      path=dataset,
                      ^^^^^^^^^^^^^
                      config_name=config,
                      ^^^^^^^^^^^^^^^^^^^
                      token=hf_token,
                      ^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
                  info = get_dataset_config_info(
                      path,
                  ...<6 lines>...
                      **config_kwargs,
                  )
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
                  raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
              datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

ZImage AnyFlow2 4-step vs Z-Image Base 32/50-step showcase

Matched 1024px outputs on identical official prompts and seeds. Z-Image Base uses CFG 4 at 32 and 50 NFE; AnyFlow2 Step-1000 uses 4 NFE with no extra inference CFG (teacher CFG 4 was distilled into the student).

Samples are ranked by AnyFlow2's positive per-image selection-score delta over the stronger of Base-32 and Base-50. This is a diagnostic showcase, not a random or representative benchmark subset. See metadata.jsonl for prompts, scores, details, and seeds.

Qwen-Image-Bench (English)

01. `qwen_image_bench_en:769:0`

Prompt: After the bent spring is released, it bounces the ball up, and the ball follows a clearly identifiable upward trajectory; the spring coil rebounds naturally with the correct direction of force, and the base is stable without penetration; the light makes the metal spring appear realistically glossy, the surface material of the ball is identifiable, and the overall image is high-definition without any ghosting or artifacts.

Selection score: Base-32 33.33, Base-50 33.33, AnyFlow2-4 63.16 (delta vs strongest Base +29.82).

Qwen uses the mean normalized score over applicable official-judge criteria; LongText uses official word accuracy.

Z-Image Base · 32 NFE · CFG 4	Z-Image Base · 50 NFE · CFG 4	AnyFlow2 Step-1000 · 4 NFE · no extra CFG

02. `qwen_image_bench_en:42:0`

Prompt: Generate a plant science popularization poster: The main theme is "Lotus (lotus)" in a retro style (9:16), strictly following the following format: left side: full plant cross-section (muddy rhizome → petiole → lotus leaf + flower), two circular anatomical close-ups in the middle (rhizome cut surface holes + lotus seedpod honeycomb), right side: four-part life cycle diagram (young leaves/flower buds/withered lotus seedpods/withered lotus in autumn water), bottom: application display (lotus root slices/lotus seed soup/lotus leaf tea); style: "Compendium of Materia Medica" × Western Flora, watercolor, old beige parchment background; top title: "Lotus", including Chinese handwritten annotations (stamens/carpels, etc.) and indicator lines, requiring clear and legible text and rigorous layout.

Selection score: Base-32 32.90, Base-50 30.34, AnyFlow2-4 60.00 (delta vs strongest Base +27.10).

Qwen uses the mean normalized score over applicable official-judge criteria; LongText uses official word accuracy.

Z-Image Base · 32 NFE · CFG 4	Z-Image Base · 50 NFE · CFG 4	AnyFlow2 Step-1000 · 4 NFE · no extra CFG

03. `qwen_image_bench_en:701:0`

Prompt: Coffee cup packaging slogan "Brewed with Soul" in a handwritten style font, letters flow naturally, curves soft, aligning with brand's warm tone.

Selection score: Base-32 35.29, Base-50 40.00, AnyFlow2-4 67.00 (delta vs strongest Base +27.00).

Qwen uses the mean normalized score over applicable official-judge criteria; LongText uses official word accuracy.

Z-Image Base · 32 NFE · CFG 4	Z-Image Base · 50 NFE · CFG 4	AnyFlow2 Step-1000 · 4 NFE · no extra CFG

04. `qwen_image_bench_en:711:0`

Prompt: At a community health post in Oaxaca, Mexico: a hand-painted wall slogan reads “Lidxi’ ya guezaa – Siempre contigo” (We are here – Always with you), blending Zapotec and Spanish; the Zapotec portion adheres to Oaxaca’s official orthography, with the glottal stop “’” in “Lidxi’” precisely marked; painted in a style by local women, using red and indigo hues drawn from traditional textiles

Selection score: Base-32 37.27, Base-50 41.67, AnyFlow2-4 67.50 (delta vs strongest Base +25.83).

Qwen uses the mean normalized score over applicable official-judge criteria; LongText uses official word accuracy.

Z-Image Base · 32 NFE · CFG 4	Z-Image Base · 50 NFE · CFG 4	AnyFlow2 Step-1000 · 4 NFE · no extra CFG

05. `qwen_image_bench_en:134:0`

Prompt: Unmelted snow lay on the black car hood, each snowflake retaining its hexagonal crystal outline and branching texture; the paint reflected ambient light to create realistic highlights, and there was a slight melting water film where the snow met the paint but did not destroy its shape; the edges were sharp, the details were rich, and there was no over-sharpening noise.

Selection score: Base-32 35.29, Base-50 40.00, AnyFlow2-4 65.56 (delta vs strongest Base +25.56).

Qwen uses the mean normalized score over applicable official-judge criteria; LongText uses official word accuracy.

Z-Image Base · 32 NFE · CFG 4	Z-Image Base · 50 NFE · CFG 4	AnyFlow2 Step-1000 · 4 NFE · no extra CFG

06. `qwen_image_bench_en:735:0`

Prompt: A storyboard illustrating a mysterious figure entering a bar, beginning with a street wide shot, followed by a close-up of the door opening, then a medium close shot at the bar, ending with the bartender’s reaction.

Selection score: Base-32 10.00, Base-50 28.70, AnyFlow2-4 53.64 (delta vs strongest Base +24.94).

Qwen uses the mean normalized score over applicable official-judge criteria; LongText uses official word accuracy.

Z-Image Base · 32 NFE · CFG 4	Z-Image Base · 50 NFE · CFG 4	AnyFlow2 Step-1000 · 4 NFE · no extra CFG

07. `qwen_image_bench_en:954:0`

Prompt: person lies in a bathtub, with rose petals floating on the surface of the water. The camera is close to the water, capturing the shoulders and collarbone.

Selection score: Base-32 30.00, Base-50 25.38, AnyFlow2-4 53.85 (delta vs strongest Base +23.85).

Qwen uses the mean normalized score over applicable official-judge criteria; LongText uses official word accuracy.

Z-Image Base · 32 NFE · CFG 4	Z-Image Base · 50 NFE · CFG 4	AnyFlow2 Step-1000 · 4 NFE · no extra CFG

08. `qwen_image_bench_en:699:0`

Prompt: The tapestry "Memory Warp and Weft" is made from scraps of old clothes, with coarse hemp thread connecting the blurry faces of three generations of grandparents and grandchildren; the fabric hangs naturally due to its own weight, and the outline moves slightly with the airflow, with the family history appearing and disappearing in the folds of the fibers.

Selection score: Base-32 17.14, Base-50 30.91, AnyFlow2-4 54.00 (delta vs strongest Base +23.09).

Qwen uses the mean normalized score over applicable official-judge criteria; LongText uses official word accuracy.

Z-Image Base · 32 NFE · CFG 4	Z-Image Base · 50 NFE · CFG 4	AnyFlow2 Step-1000 · 4 NFE · no extra CFG

09. `qwen_image_bench_en:138:0`

Prompt: Low-angle near shot of a wet asphalt street after rain at night. Puddles mirror streetlights and building façades with correct geometry; subtle ripples only, no smeared edges. Reflections reveal continuous tire tracks and gritty aggregates; avoid AI-like repeating patterns.

Selection score: Base-32 35.00, Base-50 38.75, AnyFlow2-4 61.25 (delta vs strongest Base +22.50).

Qwen uses the mean normalized score over applicable official-judge criteria; LongText uses official word accuracy.

Z-Image Base · 32 NFE · CFG 4	Z-Image Base · 50 NFE · CFG 4	AnyFlow2 Step-1000 · 4 NFE · no extra CFG

10. `qwen_image_bench_en:659:0`

Prompt: At the core of the Earth is a warm sun, whose heat makes the entire underground world habitable and full of life.

Selection score: Base-32 30.00, Base-50 30.00, AnyFlow2-4 51.82 (delta vs strongest Base +21.82).

Qwen uses the mean normalized score over applicable official-judge criteria; LongText uses official word accuracy.

Z-Image Base · 32 NFE · CFG 4	Z-Image Base · 50 NFE · CFG 4	AnyFlow2 Step-1000 · 4 NFE · no extra CFG

LongText-EN

01. `longtext_en:55:0`

Prompt: A neatly unfolded vintage city map. Its corners slightly creased from age and frequent use. At the top center, in distinct serif fonts, clearly printed text reads prominently: "Explore our Charming Streets and Landmarks!" Immediately underneath this title, in smaller italicized letters, the tagline neatly states, "A guide to the best sights, tastes, and hidden corners of Charleston, South Carolina". In the map's lower-left corner, clearly structured in a boxed legend labeled "Map Key" in uppercase letters, concise yet readable sentences explain, "Historic Sites", "Restaurants & Cafes", "Parking Areas". To the right side of the map, aligned vertically and organized neatly, a list in tidy serif typeface titled "Must-Visit Destinations" features several concise descriptions, such as: "City Market – bustling local market offering handmade crafts", and "Waterfront Park – iconic pineapple fountain". At the map's lower-right corner, clearly legible directions read, "For more information or guided tours, please visit our information booth on Meeting Street. Call us at 555-0162". Each textual section is thoughtfully laid out, comfortably spaced, and easy to follow, creating an attractive and informative visual narrative.

Selection score: Base-32 45.90, Base-50 50.82, AnyFlow2-4 86.89 (delta vs strongest Base +36.07).

Qwen uses the mean normalized score over applicable official-judge criteria; LongText uses official word accuracy.

Z-Image Base · 32 NFE · CFG 4	Z-Image Base · 50 NFE · CFG 4	AnyFlow2 Step-1000 · 4 NFE · no extra CFG

02. `longtext_en:27:3`

Prompt: A tea shop filled with organized shelves displaying jars containing various tea blends. Each tea jar has elegantly handwritten rectangular labels neatly placed at the center front; the prominently visible label reads "Summer Peach White Tea", with slightly smaller descriptive text directly underneath stating "Light floral notes with peach flavor". Surrounding jars feature similarly styled labels arranged uniformly, including "Chamomile Lavender Herbal Blend" label. The textual contents on each jar's label are carefully aligned: tea names appear in larger and emphasized handwriting on the top, and concise descriptive phrases appear neatly beneath, creating a visually appealing and easily readable layout that attracts and informs visitors clearly.

Selection score: Base-32 71.43, Base-50 71.43, AnyFlow2-4 100.00 (delta vs strongest Base +28.57).

Qwen uses the mean normalized score over applicable official-judge criteria; LongText uses official word accuracy.

Z-Image Base · 32 NFE · CFG 4	Z-Image Base · 50 NFE · CFG 4	AnyFlow2 Step-1000 · 4 NFE · no extra CFG

03. `longtext_en:68:2`

Prompt: A clean and visually engaging online travel blog interface characterized by a vibrant yet minimalistic aesthetic and intuitive layout style. Located along the very top is a neatly aligned horizontal menu bar consisting of navigation links labeled "Home", "Destinations", "Travel Tips", and "About Me", each clearly readable with subtle spacing between them. Prominently showcased at the center of the homepage, capturing instant attention, is an inspirational textual statement gracefully set in bold, serif typography that reads clearly: "Wander freely, embrace new cultures and create lifelong travel memories". Directly underneath this central textual element, the layout displays responsive rectangular thumbnails arranged neatly in rows, each linking readers to recent articles and posts. Overall, the organized structure and effective use of contrast effortlessly guides viewers toward focusing on the central, inviting textual content.

Selection score: Base-32 62.50, Base-50 62.50, AnyFlow2-4 87.50 (delta vs strongest Base +25.00).

Qwen uses the mean normalized score over applicable official-judge criteria; LongText uses official word accuracy.

Z-Image Base · 32 NFE · CFG 4	Z-Image Base · 50 NFE · CFG 4	AnyFlow2 Step-1000 · 4 NFE · no extra CFG

04. `longtext_en:62:2`

Prompt: A modern social media platform interface displayed on a sleek smartphone screen. The interface has a clean, visually appealing design with minimalistic yet vibrant icons arranged precisely at the bottom navigation bar. At the center of the screen, a prominently displayed textual post can clearly be seen, reading: "Travel far, experience more, unlock the adventure within your heart". This central phrase is set in elegant yet bold typography, enhanced with a visually balanced contrast against a subdued gradient background image depicting blurred natural landscapes. Directly above the text, neatly positioned, is a user profile image accompanied by a username along with a subtle timestamp indicating when the post was made. Beneath the main textual content are interactive buttons and icons for user engagement, distinctly arranged horizontally to indicate functionality such as "like", "comment", and "share". The overall design approach emphasizes readability and visual harmony, guiding viewers' eyes naturally toward the compelling central text.

Selection score: Base-32 76.92, Base-50 76.92, AnyFlow2-4 100.00 (delta vs strongest Base +23.08).

Qwen uses the mean normalized score over applicable official-judge criteria; LongText uses official word accuracy.

Z-Image Base · 32 NFE · CFG 4	Z-Image Base · 50 NFE · CFG 4	AnyFlow2 Step-1000 · 4 NFE · no extra CFG

05. `longtext_en:27:2`

Selection score: Base-32 71.43, Base-50 71.43, AnyFlow2-4 92.86 (delta vs strongest Base +21.43).

Qwen uses the mean normalized score over applicable official-judge criteria; LongText uses official word accuracy.

Z-Image Base · 32 NFE · CFG 4	Z-Image Base · 50 NFE · CFG 4	AnyFlow2 Step-1000 · 4 NFE · no extra CFG

06. `longtext_en:63:2`

Prompt: A sophisticated online blog homepage displayed on a bright, modern desktop screen, designed with a clear, contemporary layout. At the top of the interface, there is a visually pleasing horizontal navigation bar with neatly arranged links labeled "Home", "Lifestyle", "Wellness", and "Contact". Just below this navigation area is a featured section prominently showing textual content that reads: "Simple Steps to Help You Live Sustainably and Improve Wellbeing Daily". The main text is presented in bold and inviting serif typography to attract immediate attention, set in dark colors against a minimalistic, pastel-toned backdrop. Below this headline, organized vertically, are neatly aligned article preview cards, each containing a small preview image, article titles, author names, and publication dates. Soft, subtle separators visually structure each section, creating a harmonious design that highlights the central textual message.

Selection score: Base-32 73.33, Base-50 80.00, AnyFlow2-4 100.00 (delta vs strongest Base +20.00).

Qwen uses the mean normalized score over applicable official-judge criteria; LongText uses official word accuracy.

Z-Image Base · 32 NFE · CFG 4	Z-Image Base · 50 NFE · CFG 4	AnyFlow2 Step-1000 · 4 NFE · no extra CFG

07. `longtext_en:139:1`

Prompt: A screenshot from a weather broadcast, featuring a meteorologist standing confidently beside a colorful digital weather map with clearly illustrated temperature and precipitation forecasts. Temperatures are shown with clear numerical readings, ranging from cooler values marked in blue tones inland, gradually warming to warmer orange or red values near the coastline. Across the bottom of the frame, a semi-transparent dark-blue ticker panel stretches horizontally, carrying neat white textual subtitles stating: "Weather Alert: Tropical storm warnings issued across coastal communities, with heavy rainfall and wind gusts of up to 70 mph expected within the next 48 hours. Residents are advised to stay informed and prepare for possible evacuations". Positioned at the lower-left of this panel, the red bold capital letters "LIVE" appear clearly alongside the weather channel logo, reinforcing the urgent nature of the live weather update.

Selection score: Base-32 63.16, Base-50 81.58, AnyFlow2-4 94.74 (delta vs strongest Base +13.16).

Qwen uses the mean normalized score over applicable official-judge criteria; LongText uses official word accuracy.

Z-Image Base · 32 NFE · CFG 4	Z-Image Base · 50 NFE · CFG 4	AnyFlow2 Step-1000 · 4 NFE · no extra CFG

08. `longtext_en:120:1`

Prompt: A screenshot captured from a documentary film scene showing an aerial view of a glacier melting rapidly into the ocean. At the bottom center of the frame, a translucent dark gray rectangular caption box overlays the footage, containing the textual caption in clean white sans-serif fonts. Within this box, the text "Arctic glaciers disappearing rapidly as global temperatures reach record highs" appears aligned centrally, stretching horizontally across most of the lower portion of the screen. The caption is concise, clear, providing immediate context for viewers. The scene timestamp "00:47:23" is situated in a small white font, placed neatly at the lower-right corner, slightly below and to the right of the main caption box.

Selection score: Base-32 90.91, Base-50 90.91, AnyFlow2-4 100.00 (delta vs strongest Base +9.09).

Qwen uses the mean normalized score over applicable official-judge criteria; LongText uses official word accuracy.

Z-Image Base · 32 NFE · CFG 4	Z-Image Base · 50 NFE · CFG 4	AnyFlow2 Step-1000 · 4 NFE · no extra CFG

09. `longtext_en:34:3`

Prompt: An inviting and meticulously organized botanical garden exhibit featuring a variety of beautiful plants clearly identified by neatly printed informational labels. Prominently positioned at the center is a rectangular, easily readable tag stating, "Rare Orchid Species Preserve carefully and avoid direct sunlight exposure", highlighting critical plant care instructions. Surrounding plants each carry smaller, carefully arranged labels with concise yet informative descriptions such as "Succulents Low Water Need", "Tropical Fern Thrives in Humidity", and "Alpine Moss Cold Compatible", thoughtfully presented for visitors. All textual tags are clearly legible, uniformly styled, and professionally printed, further enhancing the educational ambiance and emphasizing the garden's dedication to providing precise plant care knowledge in a clear, accessible manner.

Selection score: Base-32 78.26, Base-50 82.61, AnyFlow2-4 91.30 (delta vs strongest Base +8.70).

Qwen uses the mean normalized score over applicable official-judge criteria; LongText uses official word accuracy.

Z-Image Base · 32 NFE · CFG 4	Z-Image Base · 50 NFE · CFG 4	AnyFlow2 Step-1000 · 4 NFE · no extra CFG

10. `longtext_en:27:0`

Selection score: Base-32 57.14, Base-50 35.71, AnyFlow2-4 64.29 (delta vs strongest Base +7.14).

Qwen uses the mean normalized score over applicable official-judge criteria; LongText uses official word accuracy.

Z-Image Base · 32 NFE · CFG 4	Z-Image Base · 50 NFE · CFG 4	AnyFlow2 Step-1000 · 4 NFE · no extra CFG

Downloads last month: 317

Total file size:

91.4 MB

ZImage AnyFlow2 4-step vs Z-Image Base 32/50-step showcase

Qwen-Image-Bench (English)

01. qwen_image_bench_en:769:0

02. qwen_image_bench_en:42:0

03. qwen_image_bench_en:701:0

04. qwen_image_bench_en:711:0

05. qwen_image_bench_en:134:0

06. qwen_image_bench_en:735:0

07. qwen_image_bench_en:954:0

08. qwen_image_bench_en:699:0

09. qwen_image_bench_en:138:0

10. qwen_image_bench_en:659:0

LongText-EN

01. longtext_en:55:0

02. longtext_en:27:3

03. longtext_en:68:2

04. longtext_en:62:2

05. longtext_en:27:2

06. longtext_en:63:2

07. longtext_en:139:1

08. longtext_en:120:1

09. longtext_en:34:3

10. longtext_en:27:0

01. `qwen_image_bench_en:769:0`

02. `qwen_image_bench_en:42:0`

03. `qwen_image_bench_en:701:0`

04. `qwen_image_bench_en:711:0`

05. `qwen_image_bench_en:134:0`

06. `qwen_image_bench_en:735:0`

07. `qwen_image_bench_en:954:0`

08. `qwen_image_bench_en:699:0`

09. `qwen_image_bench_en:138:0`

10. `qwen_image_bench_en:659:0`

01. `longtext_en:55:0`

02. `longtext_en:27:3`

03. `longtext_en:68:2`

04. `longtext_en:62:2`

05. `longtext_en:27:2`

06. `longtext_en:63:2`

07. `longtext_en:139:1`

08. `longtext_en:120:1`

09. `longtext_en:34:3`

10. `longtext_en:27:0`