Task
The client wanted to test demand for AI narrator videos and the sales funnel around the service. The MVP needed to fit a limited budget and let users complete the workflow inside Telegram without installing a separate app.
Solution
I built a Mini App with Vue and Django. Users choose a presenter, background, placement and voice, add text and submit the video for generation. The app includes a plan, remaining minutes and a history of results.
The working product generates videos through the CDNVideo API. Celery starts the job and checks its status while the user stays in the app. Each video record keeps its text and a copy of the selected settings.
The demo lets visitors complete the editor, build a mockup, save it to history and check minute accounting. AI generation is simulated.
What was done
- Launched an MVP as a Telegram Mini App
- Built an editor for presenter, background, placement, voice and text
- Integrated video generation through the CDNVideo API and Celery background jobs
- Connected video settings, status and history in the app
- Added plans and available generation time accounting
Result
More than 200 videos have been created in the working app, with about 50 users per month. These figures let the client test demand with real users and decide how to develop the product.
These are usage figures from the working app. The demo contains a small set of fictional records and demonstrates the workflows without account access or paid generation. Expansion to MAX remains a development plan, rather than a completed part of the project.
Developer's note
What was hard
The editor and a long generation job need consistent state. In the original app, the settings are copied to a separate VideoJob; Celery starts processing and checks its status. Changes to the next draft must not alter a video already submitted.
I kept that principle in the demo. Each result has its own settings, minutes are reserved while the mockup is built, and reset cancels an unfinished operation and clears its state.
What came easy
The user workflow divides naturally into five clear steps. The presenter, background and placement appear immediately in the composition; voice samples can be played, and text and speed checked before submission. This makes the product easy to try without explaining the internal APIs.
For the demo, I prepared separate illustrations and speech samples. They let visitors check selection and voice playback without a paid service.
Technologies
PythonDjangoVue