- published_prompts
- Rows of the PicGens catalog that are published (is_published = true) on the edition date, after takedowns and blocked creators are removed.
- model_share
- Published prompts carrying a stored model label divided by all published prompts. "Unlabelled model" is a visible bucket; no model is inferred from prompt wording.
- category_share
- Published prompts carrying a category tag divided by all published prompts. Tags are assigned from prompt-text keywords and can overlap, so shares do not sum to 100%.
- source_month
- Calendar month (UTC) of the source post timestamp. Rows without a source-post timestamp (catalogued templates) are excluded from this dimension.
- prompt_rows
- Published prompts with non-empty prompt text: the denominator for every prompt-anatomy share.
- length_chars
- Prompt length in Unicode code points after trimming surrounding whitespace.
- length_words
- Number of whitespace-separated tokens after trimming. Chinese, Japanese and Korean text has few spaces, so word counts understate those prompts; use characters to compare across languages.
- structured_json
- The whole prompt (after trimming and removing an optional Markdown code fence) parses as a JSON object or array.
- template_slots
- The prompt contains at least one fill-in slot: [text] or {{text}} of 1–60 characters with at least one letter, no quotation marks or line breaks, and not followed by "(" (a Markdown link).
- in_image_text
- The prompt asks for visible text in the image: it uses a text-rendering term (text, lettering, typography, typeface, font, headline, caption, subtitle, slogan, tagline, wordmark, logotype, signage, "the words"; 文字/字样/标题/文案/テキスト/텍스트) or introduces a quoted string with says/reads/title/label/sign/banner — after removing negations such as "no text", "without lettering" or "text-free".
- aspect_ratio_mention
- The prompt names an explicit aspect ratio W:H from this list: 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9, 9:21, 2:1, 1:2 (e.g. "--ar 16:9" or "9:16 framing"), not preceded or followed by another digit, dot or colon.
- camera_spec
- The prompt names a camera setting: a focal length (e.g. 85mm), an aperture (f/1.8), an ISO value (ISO 200) or a shutter speed (1/250s).
- cjk_script
- At least 30% of the letters in the prompt are Chinese, Japanese or Korean characters.
- image_orientation
- From the stored width and height of the first published image: portrait (height > width × 1.05), landscape (width > height × 1.05), otherwise square. Rows without stored dimensions are excluded.
- color_tags
- Up to three dominant-colour buckets per image, extracted from the thumbnail palette. Share = images carrying the colour divided by images with any colour tag.
- engagement
- Likes, bookmarks and views of the source post as stored at the last stats refresh, for rows with a source-post timestamp. Views are counted only where a non-zero view count was stored. Medians, not means, because counts are heavily skewed.