Dataset edition · published September 10, 2026 · schema v1
AI Image Prompt Statistics: 2026-09-10 Edition
This is an archived edition, kept at a stable URL so earlier citations stay valid. The current edition is 2026-10-04; the statistics hub always shows the latest figures.
This edition describes the published PicGens prompt corpus as of 2026-09-10: 3,698 public creator prompts, counted by stored model label, keyword-assigned category tag and source-post month. It is a first-party corpus description — not search demand, rankings, market share or a measure of model quality.
1.The PicGens AI image prompt corpus held 3,698 published prompts on September 10, 2026. (PicGens, 2026-09-10)
2.GPT Image 2 is the AI image model creators share the most prompts for: 56.6% of published prompts (2,093 of 3,698). (PicGens, 2026-09-10)
3.Nano Banana Pro ranks second, with 12.6% of published prompts (466). (PicGens, 2026-09-10)
4.The Nano Banana family (Nano Banana, Nano Banana 2 and Nano Banana Pro) accounts for 22.9% of published prompts (847). (PicGens, 2026-09-10)
5.Midjourney accounts for 3.1% of published prompts (115). (PicGens, 2026-09-10)
6.12.4% of published prompts (457) carry no model label. (PicGens, 2026-09-10)
7.The corpus grew 14.2% since the 2026-08-15 edition, from 3,239 to 3,698 published prompts. (PicGens, 2026-09-10)
8.GPT Image 2 added 457 prompts since the 2026-08-15 edition, moving its share from 50.5% to 56.6%. (PicGens, 2026-09-10)
9.Portraits is the most common use case: 56.4% of published prompts carry the Portraits tag (2,086). (PicGens, 2026-09-10)
10.Posters & visuals (42.6%) and Illustration & 3D (34.5%) are the next most common use-case tags. (PicGens, 2026-09-10)
11.Ads & product follows, tagged on 23.5% of published prompts (868). (PicGens, 2026-09-10)
12.12.6% of published prompts (467) are reusable prompt templates (Prompt templates tag). (PicGens, 2026-09-10)
13.June 2026 was the busiest source month, with 1,259 prompts (39.0% of the 3,231 prompts with a source-post date). (PicGens, 2026-09-10)
Every figure is computed from PicGens’ own published prompt corpus and can be reproduced from the CSV or JSON of this edition. Shares use all published prompts as the denominator unless stated otherwise.
How to cite these statistics
Cite the dated edition, not this page’s latest view: the edition URL and its files never change, so a quoted number stays verifiable after the corpus grows.
Suggested citation
PicGens. (2026, September 10). AI Image Prompt Statistics 2026 (edition 2026-09-10) [Data set]. https://www.picgens.com/research/prompt-trends/2026-09-10
Short attribution
Source: PicGens, AI Image Prompt Statistics 2026, edition 2026-09-10 (https://www.picgens.com/research/prompt-trends/2026-09-10)
Rows where is_published = true on the edition date.
Model labels
10
“Unlabelled model” stays visible; no model is inferred from prompt wording.
Latest ingest
2026-09-10
Latest represented source date: 2026-08-14.
Published prompts by stored model label
GPT Image 2 is the largest group in this edition, with 2,093 prompts (56.6%). These are corpus counts only: they do not measure which model is best, most used overall or more likely to reproduce a result.
Model label counts in the 2026-09-10 PicGens edition
Stored key
Label
Published prompts
Share
gpt-image-2
GPT Image 2
2,093
56.6%
nanobanana-pro
Nano Banana Pro
466
12.6%
unknown
Unlabelled model
457
12.4%
nanobanana-2
Nano Banana 2
286
7.7%
midjourney
Midjourney
115
3.1%
nanobanana
Nano Banana
95
2.6%
seedream
Seedream
89
2.4%
seedance
Seedance
57
1.5%
gpt-image-1
GPT Image 1
39
1.1%
veo
Veo
1
0.0%
Published prompts by category tag
Category tags are assigned from prompt-text keywords. A prompt can carry more than one tag, so category totals intentionally do not add up to the corpus total.
Category tag shares in the 2026-09-10 edition
Portraits56.4% (2,086)
Posters & visuals42.6% (1,574)
Illustration & 3D34.5% (1,275)
Ads & product23.5% (868)
Prompt templates12.6% (467)
Brand & logo10.1% (373)
Video7.8% (288)
Wallpaper4.7% (174)
UI & interfaces1.8% (65)
Charts & infographics1.7% (63)
Published prompts by source-post month
The month dimension uses the stored source-post timestamp, not the ingest date. It covers the 3,231 prompts that carry a source-post date, so it does not sum to the corpus total. It describes when the represented public examples were posted; it is not a time series of demand or traffic.
Source-post month counts in the 2026-09-10 PicGens edition
Month
Prompts
2023-10
2
2025-04
2
2025-05
1
2025-07
4
2025-08
1
2025-09
3
2025-10
1
2025-11
7
2025-12
39
2026-01
146
2026-02
113
2026-03
59
2026-04
88
2026-05
185
2026-06
1,259
2026-07
1,214
2026-08
107
Method
Read published rows only and aggregate by model, existing category tag and tweet-created month; the 2026-09-10 run read 3,698 rows through a paginated Supabase query.
Keep “unknown” as a visible bucket instead of inferring a model from prompt wording.
Do not expose raw prompt contents, row identifiers, creator handles, private fields or per-row behaviour.
Counts are a dated corpus snapshot; they are not search volume, ranking, engagement or market share.
Copy rate: unavailable — no copy event export was authorised for this snapshot.
Limitations
The latest ingested row is 2026-09-10 and the latest source-post date represented is 2026-08-14; this is not a live trend feed.
Category tags can overlap, so category totals are not expected to sum to the corpus total.
This release adds 467 catalogued prompt-template rows that carry no source-post timestamp, so the month table covers 3,231 of the 3,698 published rows.
No GSC, GA4, CrUX, backlink or copy-rate data was available or inferred.
No raw prompt text, row IDs, creator handles, source URLs, private fields or per-row behaviour are distributed with this edition.
Metric definitions
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.
Agents and researchers can use the JSON or CSV distribution as a stable, dated citation target. The files hold aggregate dimensions; they do not expose individual creator content or grant rights to reuse any source image or prompt.