How people in India create images and videos with AI tools, and where those tools fall short in representing local culture, language and context.
This study explored how people create images and videos with AI tools, and where those tools fall short in representing local culture, language and context. It looked at the challenges creators face, and what they expect from authentic results that they can trust.
The aim: insights to guide an image and video generation experience that feels accurate, culturally respectful and empowering for local creators. The client asked two questions:
50 participants with a mix of six languages: English and five regional languages. Everyone had the same task: create an image that represents a local vegetable market in their area.
In their own languageHindi and English sessions were moderated in-house, by my co-researcher. For Telugu, Bengali, Marathi and Tamil, we hired moderators who spoke each language, so participants could talk freely. A translator interpreted each session into English live, and we took notes from the translation.
Two phases of 25: first a mix of students, working professionals and homemakers, then content creators and people who use video generation in their professional lives.
"Generate an image of a vegetable market that looks local to you, with all the culturally relevant elements."
Before the interview, each person made two images on two different AI tools. So we walked in already knowing what "local" meant to them. In the interview, a probe of images we had created asked: does it look AI-generated, what's missing, and what looks too AI?
We did sense-making for all 50 participants, using the same structured sheets, so patterns could surface across languages and cities.
For each step of each person's journey, we looked at what they did and thought, and how it felt:
Here is one of them, from the moment someone needs an image to the moment they use it.
Scroll sideways to see the whole journey →
It goes beyond geography to cultural familiarity and a sense of belonging. Most people saw local imagery as an authentic reflection of their own surroundings, and every city pictured it differently. A few linked "local" to rural settings or low-quality images.
Across cities, participants equated “local feel” with how people look, move, and behave, not just the setting. People and their faces, expressions, attire, gestures, postures and visible interactions were essential indicators of authenticity. Small details like someone paying a vendor, a crowd in motion or, sweat on people’s faces added to this.
Participants repeatedly pointed out misrepresented or unrealistic elements like non-Indian faces, mismatched or similar clothing, sparse crowds, and cartoonish visuals. Such inaccuracies instantly reduced the authenticity of an image.
In a nutshell, image resolution and clarity is the most important visual parameter when generating images, since clearer pictures most closely resemble real (non-AI) pictures.
"Such in-depth detailing and clarity of the faces, and even of the background, should be there that it is not blurry on zooming in."
Even when physical landmarks were missing, participants looked for emotional familiarity within images. They wanted visuals with atmospheres that felt like their city, not just ones that looked like it.
Recognizable streets, markets, religious sites, or iconic buildings helped anchor the identity and location of a place. Even minor inaccuracies in these elements disrupted the sense of authenticity. They wanted these details to ground the image, rather than stereotyping it.
We broke "local and cultural relevance" down into what an image or video has to get right in the Indian context: four pillars, and what each one encompasses.
Recognizable places of importance or iconic landmarks from their context. Also, the characteristics of their place and how it is perceived by the locals themselves and not from an outsider lens
A sense of how many people would be present in a place at a given point in time
Customs and traditions from their region and culture to be depicted in the image in a way that it truly represents their place.
Social code about what is written and said, what is worn – when and how, how one behaves in a specific festival
Show how people would interact with each other in the context of the place, time, relationship amongst people, and occasion specific.
Activities and interactions should be shown
Capture how people in their context think emote.
The inner thinking of people should be depicted
People’s facial tonalities and expressions need to be shown, along with their complexion and attire which represents their context
What is written, how is it written, in what language – use of colloquial words and local languages etc.
Text should not be jumbled but in a language which feels realistic and natural in the scene depicted
Participants were highly attentive to details of religious context, even subtle deviations were perceived as inauthentic
Cultural symbols, like the Golden Temple, were seen as sacred and must never be shown inaccurately. Participants expected such cultural taboos to be understood to avoid hurtful representations
Most people typed rather than spoke their prompts, so they could compose their thoughts, check them and cover every requirement. Voice was for when they were in a hurry, and felt unreliable.
English was the main prompting language. Some tried Hindi, Tamil, Tanglish or their regional language, but switched back to English for a good result:
ChatGPT, Gemini, Meta AI, Perplexity, Canva AI, Envato and Higgsfield were the major tools used for image and video generation. Here is how the two main tools compared:
Beyond the two main tools, the choice came down to convenience, speed and accuracy. For video, no single tool felt seamless, so people switched tools job by job.
The workaroundMany people split the job: ChatGPT to put their idea into words, then Gemini to turn that prompt into the image.
The insights and findings we submitted helped the client make more informed decisions, across three areas:
This was my first project at Peepal Design where I took part from end to end, across almost every process. It was a very extensive study, with a huge sample of 50 participants and a wide variety of people, from people who generate images to video creators who use video generation heavily in their professional lives.
I coordinated with the field recruiters, and with four regional moderators. Since they only ran the sessions in their own language, we shared the probes and activities with them, walked them through the discussion guide, and trained them before the sessions, including how we ran the activities on Miro to gather views on the images. All of this was new to me, and I learned a lot about coordination.
Journey mapping taught me to write what isn't visible on the surface: to think hard about what people do and feel, and bring it out. This project shaped much of the way I write at Peepal Design.