( Services )
Custom AI App Development
I build AI web apps from idea to payments and maintenance. Four of my own run under real traffic: more than 90,000 users and over 300,000 generated images.
A custom AI app is a web tool built around an artificial intelligence model that generates images, video, voice or text, together with everything a real product needs: accounts, payments and maintenance. I build them from $8,000, and every project starts with a $790 workshop that I deduct from the build price.
What I have actually built
This is not a theoretical offer. Four of my AI apps are live, take payments and have users:
- PixiFixi: an AI photo editor with more than 60,000 users and over 300,000 generated images
- OgroVision: AI garden visualization, more than 30,000 users
- IzbaViz: AI interior visualization, including animated versions
- Lovelio: AI-generated wishes, cards, songs and video greetings
Each one went from idea through launch to maintenance under real traffic. That matters when you compare it with a contractor who hands over the project and disappears before anyone sees how it behaves at a thousand generations a day.
If you want one of these tools on your own site, such as a virtual try-on in your store or interior visualization for a furniture showroom, you do not have to build it from scratch. I describe that on the AI virtual try-on and visualization page.
What building them taught me
The hardest part of an AI app is not the model. Everyone has access to good models now. The difficulty is everything around them: what to show when the model returns an error or nonsense, how to stay within a cost per user, what to do when a provider raises prices or retires a model version.
One provider is not enough. In PixiFixi every tool has a backup model with a different provider, and the app switches to it on its own when the main one stops responding. Users never notice. They only see that it works.
The interface decides everything. I design for people who do not know what a prompt is and should not have to. You upload a photo, pick a ready-made option, get the result.
Model costs have to be worked out before the first line of code. An app that costs pennies with a hundred users can eat the whole margin at ten thousand. Before shipping a new feature I run a test on real examples and record the cost of every attempt. That is how the animations in IzbaViz came about: the video feature went live only after a 26-clip test.
When a custom app makes sense
It does when no off-the-shelf tool does what you need and the process repeats often enough for the savings to add up. Or when you want a product you earn money on, not just an internal improvement.
It does not when there is a subscription tool that already does 80% of what you want. A custom app means a build cost plus maintenance. I do sometimes talk people out of building and point them to an existing product.
It also does not when you do not yet know what the app should do. In that case it is cheaper to start with automation on existing tools, test your assumptions and only then build.
How a project runs
- Workshop ($790). We agree the scope, I test models on your examples and work out the cost per user. You get a written recommendation, including when it says "let us not build this".
- First working version. One flow, from user input to result and payment. Usually 4–8 weeks.
- Building on a working foundation. More tools, more languages, an Android app.
- Maintenance. Model updates, error and cost monitoring: the same work I do every day on my own apps.
Working with me from abroad
Devlom is a one-person studio in Poland. You talk directly to the person who designs and writes the app, not to an account manager. I am on Central European Time, which overlaps with UK and EU working hours and with the US morning. Communication runs over email and video calls.
Your data
Before starting we decide whether data may go to an external model provider, whether a data processing agreement is needed, and whether a model hosted on your own infrastructure is an option. I work under EU data protection rules (GDPR) by default. This affects both the architecture and the cost, which is why we discuss it at the beginning and not after launch.
What the apps I have already built can do
Image
Photo editing and retouching, restoring old photos, background removal, upscaling, virtual try-on, interior and garden visualizations with zones the user marks by hand.
Video and voice
Bringing photos to life, animated visualizations, talking avatars, voiceover and songs generated from a few sentences the user types in.
Everything around the model
Accounts, credits, payments, invoices, content moderation, language versions and automatic fallback to a backup provider when the main one stops responding.
My own AI apps: live, earning, with real users
PixiFixi — AI photo editor
pixifixi.app- More than 60,000 users, around 1,000 generations a day
- A dozen or so tools: retouching, restoration, try-on, photo shoot, animation
- Several models from several providers with automatic fallback
- Stripe and PayPal payments, invoices issued automatically
OgroVision — AI garden design
aigarden.design- More than 30,000 users, English and Polish versions
- Ready-made styles or zones drawn on the photo with a description
- The same garden across the seasons
- Live since February 2025
IzbaViz — AI interior visualization
izbaviz.com- A photo of a room turned into a design in the chosen style
- Animated visualizations (image to video)
- Video feature shipped after a 26-clip test with costs
- English and Polish versions
Lovelio — AI wishes and cards
aiwishes.app- Personalized wishes and greeting cards
- An AI-generated song with the wishes
- Video wishes with a talking character
- English and Polish versions
How much does an AI app cost?
AI workshop
$790
- 3 hours on your process or idea
- Whether AI makes economic sense here
- Estimated model cost per user
- Written recommendation and a price range for the build
- Deducted from the build price
Custom AI app
from $8,000
- First working version in 4–8 weeks
- Accounts, payments and an admin panel
- Several model providers with fallback
- AI model costs billed separately, by usage
- Maintenance and development after launch
Tools I use
- OpenAI (GPT, gpt-image)
- Anthropic Claude
- Google Gemini and nano-banana
- Flux (Black Forest Labs)
- Seedream and Seedance (ByteDance)
- Replicate
- fal.ai
- Laravel
- React
- Python
- Stripe, PayPal
Frequently asked questions
A simple web app with one main flow (say, the user uploads a photo, picks a variant, pays and downloads the result) starts at $8,000 with me. On top of that comes the cost of the models themselves, billed by usage, usually between a cent and a few tens of cents per generated image. I give an exact quote after the workshop, once I know how many flows, integrations and languages you need.
A first working version you can show to users usually takes 4–8 weeks. I do not start with the full feature list. I start with one flow that has to prove two things: that the idea works, and that the model cost fits inside what the user will pay.
The one that gives good enough quality at a cost that adds up, and never just one. In PixiFixi every tool has a main model and a backup from a different provider, because models get retired, change their prices and go down. I choose based on a test on your own examples, not on a leaderboard.
Yes. All four of my apps sell credits or packs through Stripe and PayPal, and invoices are issued automatically. We settle the payment model (credits, subscription, one-off purchase) in the workshop, because it decides whether the app earns enough to cover the model costs.
If we use a model through its API, yes. That is why before starting we decide which data may leave your infrastructure and whether a data processing agreement is needed. API providers (OpenAI, Anthropic, Google) do not train models on API data by default, but this should be a conscious decision made up front, not after launch.