Who are the darktable users in 2022 ?

In 2020, I surveyed the darktable user base and published the results¹ : a population of male engineers and scientists, more educated and vastly more computer-literate than the general population, with close to no women and few professional photographers. The conclusion was blunt : a photo processing software had managed to attract everybody but the people who make photography their trade.

I ran the survey again in October 2022, with a larger questionnaire. Life decided otherwise for the processing : the data sat in a CSV file for four years. Here are, at last, the results — n = 1253, which makes it the larger of the two samples. Where the questions match, I compare with 2020.

Methodology

The survey was conducted in English on poll-maker.com, with 44 questions (2020 : 26). 1253 respondents participated ; the survey recorded IP addresses to prevent multiple takes, and no duplicate respondent identifiers are present in the data.

Improvements over 2020 :

• the native-English question is now asked directly instead of being inferred from IP addresses : 373 native speakers, 880 non-native ;
• a gender question was added — the 2020 edition did not poll it, and estimated the share of women indirectly ;
• explicit questions on operating system, graphic tablets, and MIDI controllers ;
• a contributor question (“Have you contributed to darktable source code ?”), allowing a users-vs-contributors split ;
• per-area satisfaction questions (culling, batch editing, reconstruction, colour, tonal, export, library).

Sampling caveat, stated plainly : 68 % of respondents found the survey link on a channel dedicated to darktable, 11 % on FLOSS-photography channels, 11 % on general photography channels. This sample is therefore the engaged community, not the silent user base — the same caveat applies to 2020, and if anything the bias works against the conclusions below : the engaged community is the most favorable sample the project could hope for. Only 5.9 % of respondents report having taken the 2020 survey, so the two samples are essentially independent.

Processing was done with the same Python/Pandas/Plotly pipeline as in 2020.

Results

User’s origin

Same picture as 2020 : Western Europe and North America dominate, with Germany (206) and France (202) far ahead of the United States (176). The historical proximity between the development team and the German- and French-speaking communities remains visible in the engagement.

Gender

The 2020 edition estimated women at less than 2 % of the user base, from forum and YouTube audience data. The 2022 survey asked directly : “What is your gender ? (as written on your passport)”.

25 women out of 1243 answers : 2.0 %. The indirect estimate was exact.

For calibration : women are 49.3 % of professional US photographers², 75 % of US photography graduates³, and 63 % of French photography graduates⁴. The trade is majority-women ; the tool’s community is 2 % women. That is not a gap, it is a wall — a 25-to-30-fold under-representation, measured.

Among the 65 respondents who report having contributed to darktable source code, 2 are women.

Age

The distribution is essentially stable, with a slight shift : the 26-35 class grew from 17.0 to 21.2 %, and the median moved from the [46;55] class to the upper edge of [36;45]. The share of respondents above 56 is unchanged (27.1 % → 26.8 %) — more than a quarter of the user base. The 2020 conclusion stands with a minor correction : the archetypal darktable user is a middle-aged male engineer, and the community did not get younger in any way that matters ; the modal user is still in his forties, twenty years older than the modal FLOSS contributor measured by Lakhani & Wolf in 2002⁵.

Comparing against the age distribution of US internet users (US Census 2021 single-year estimates weighted by Pew internet-use ratios, as in 2020) :

The [36 ; 65] range is over-represented by 17.2 points relative to US internet users — in 2020, the excess was 13.6 points. The middle-aged bulge did not shrink ; it grew.

Splitting by contribution status :

Education and background

53.4 % of respondents hold at least a master’s degree — against 14.4 % in the general US population⁶.

Restricting to US respondents aged over 25, to compare against the US Census educational-attainment baseline as in 2020 :

33.9 % of US respondents (25+) hold at least a master’s degree, against 14.4 % of the US population — practically identical to the 2020 figure (35 %). The education bias is stable across editions.

The 2022 survey split the studying-background question into finer categories than 2020, with a telling result :

Computer science or electrical engineering is, alone, the largest field : 44.9 % of respondents. All sciences together : 78.5 %. All arts together : 11.3 %, of which graphic arts 7.6 %. In 2020 I wrote that it was concerning for a photo editing application to attract an absolute majority of scientists and an anecdotal minority of art graduates ; in 2022, with a finer measurement, the single most common formal training among darktable users turns out to be programming itself.

Crossing background with education level and age, for both groups (densities in % of the group’s respondents) :

Computer skills

The same skill ladder as in 2020 was used. The levels shown in the graph correspond to the following statements, of which respondents picked the highest that applied :

• 0. No admin skills — “Not much. Someone handles the computer administration for me.”
• 1. Installs, needs help on errors — “I know how to install/update software, but I need help when error messages appear.”
• 2. Solves errors, no command line — “I know how to solve errors as long as no console command line is required.”
• 3. Power user (shell) — “I know how to use the shell and command line to access system configuration.”
• 4. High-level programmer — “I know how to write a script to loop over files to apply batch operations.”
• 5. Low-level programmer — “I know the output to ( (int)x << 1 == (int)x * 2 )” (a C bitshift).

As a baseline, the reference distribution of computing skills in the general population, from the same ITU data as 2020 :

54.6 % of respondents are programmers (levels 4-5). In 2020, this was 46.3 %.

Crossing skills with background and education :

47.3 % of native and 54.4 % of non-native respondents are simultaneously science-trained and programmer-level — in 2020, that joint density was 36 to 39 %. The core of the user base is not merely technical-adjacent ; it is the intersection of scientific training and programming skill, and that intersection grew by about 10 points in two years.

The within-class view (how skills distribute inside each background) shows the same asymmetry as 2020 — programmers are over-represented in every field, including the arts :

The computer-literacy filter did not weaken between 2020 and 2022 — it tightened, by 8 points. Whatever the project shipped in those two years selected the surviving user base further toward programmers.

Operating systems and input devices

Linux 65.7 %, Windows 39.1 %, Mac 11.6 % (multi-answer). The 2020 finding is unchanged : the programmer bias is not a Linux artifact, as the OS × skills views confirm :

The gear questions settle a design-priority debate with numbers :

• MIDI controllers : 10.0 % own one ; 2.4 % use it with darktable ; 95.2 % report never interacting with darktable through one.
• Graphic tablets : 37.4 % own one ; 6.0 % use it with darktable ; 83.0 % never interact through one — for a device that half the photo-retouching trade would call essential, the gap between ownership (37 %) and use (6 %) says something about how well the software supports it.

In 2021, darktable’s keyboard-shortcut system was rewritten — tripling its code size and breaking basic keyboard behaviors in the process — largely to support MIDI controllers. The 2022 numbers are the epitaph of that trade-off : the rewrite served the 2.4 %, at a maintenance and usability cost carried by the 100 %, while the four-times-larger tablet constituency remained poorly served.

Relation to photography

Photography is a hobby or passion for the overwhelming majority ; 9.2 % report any professional practice (part-time or full-time). 69.5 % learned photography at least partly alone on the internet ; 23.2 % have had formal training of any kind. The median yearly photography spending (optional question, n = 646 after outlier removal) is 500 USD, with a 90th percentile at 2000 USD.

Crossing learning and endeavour with background, and with each other :

Keywords

As in 2020, respondents picked the words describing their photography best (multi-answer) :

“Personal”, “artistic” and “creative” remain the leading self-descriptions, consistent with the hobbyist profile — as in 2020.

Satisfaction by workflow area

New in 2022 : respondents rated how much they like darktable’s handling of each workflow area.

The detailed per-area numbers deserve their own analysis (a follow-up article), but one caution applies to all of them, drawn from the 2020 study : a satisfaction survey on a self-selected population of survivors measures who stayed, not how good the product is. The people whom the culling or library workflows drove away are, by construction, not in this sample to rate them.

Conclusion

*The education question changed wording between editions (“education level” → “highest education level”) ; treat the 2020→2022 comparison as indicative.

In 2020, I concluded : “this is bad.” Two years later, with a bigger sample and finer questions : it is not getting better, it is getting purer. The share of programmers rose by 8 points. The single largest formal background is computer science. The women are still absent — now measured rather than estimated, 2.0 % against a trade that is half to three-quarters women. The gear that ships with the trade (tablets) is owned by a third and used by a twentieth ; the gear beloved of tinkerers (MIDI) drove a shortcut-system rewrite for 2.4 % of users.

A software project reveals who it is for by who it keeps. darktable’s user base is not a random sample of photographers that happens to skew technical ; it is a population filtered by the software into homogeneity — and the filter is tightening. “Created for photographers, by photographers”, says the project. The data says : created by engineers, kept by engineers.

Data & code availability

• Anonymized 2022 survey data (n = 1253 ; IP addresses, city/region, timestamps, browser tags and email fields removed) : darktable-survey-2022-anonymized.csv
• Processing code (Python / Pandas / Plotly) : process-survey-2022.py
• For the 2020 edition¹, published under the same transparency terms : darktable-survey-2020-anonymized.csv and process-survey-2020.ipynb

References