Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitchesCalling a video model was the least difficult part of building Dantiva, an AI photo-to-video service sold through a website, a Telegram bot and a Telegram Mini App. The harder work sat around that call: accounts, one wallet shared by all three entry points, a credit ledger that handles failed jobs, a support and refund console, and a provider content policy that blocked the exact use case the product was built on.
That is the account Maksim Ilin, an AI engineer and consultant, published on September 19, 2026, a day after Dantiva’s subscriptions launched. What follows separates what he reports from what he has not yet shown, and includes the figures he gave with their attribution.
As an Amazon Associate I earn from qualifying purchases.
What Dantiva does
A user picks a template, uploads a photo and receives an eight-second clip with sound. According to Ilin’s article, video generation used Google Veo 3.1 Fast or Lite. Gemini image models generated and edited still pictures, and a separate Gemini model rewrote short user prompts before they reached the video model.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →The service runs on three entry points: a website, a Telegram bot and a Telegram Mini App. All three share one account and one token balance. Around the generation step, the product includes a results library, projects, a legal center in two languages, and an admin panel that support staff use for refunds.
#1 Best Overall
Where the effort actually went
Ilin’s first bot was a Cloud Run deployment that made a single Veo API call and kept its state in memory. He describes it as a working demo that was not ready to sell. The gap between that demo and a sellable product was not one missing feature. It was a stack of systems that the demo did not need: accounts, balances, templates, support tooling and administration. In his telling, those systems consumed most of the build.
That is the core of his thesis. Model access was comparatively easy to obtain. Identity, shared state, payment correctness, support and policy fit were where the hard decisions were.
One wallet across three front doors
Each surface identifies users differently, and each purchase has to land in the same wallet regardless of where it was made.
| Surface | How the user is identified | What it means for the build |
|---|---|---|
| Website | Google sign-in or email verification | Needs its own account lifecycle and verification flow |
| Telegram bot | Telegram identity | Identity comes from the messaging platform, not from a sign-up form |
| Telegram Mini App | A signed Telegram payload received by the app | The server must verify the signature before trusting the user |
Ilin says the three interfaces drifted apart during development. He added a parity check across them after finding a feature that worked in one surface and not another. The design goal was simple to state and tedious to keep: one balance, one set of rules, and any screen that shows a number should be reading that same number.
Treat generation credits as a ledger
Ilin describes token balances as financial transactions, not counters. The flow he reports works like this:
- Reserve. When a generation job starts, the tokens it will cost are reserved against the balance.
- Capture. When a result arrives, the reserved tokens are captured as spent.
- Refund. If the provider fails, the reserved tokens are returned to the wallet.
Three guards sit alongside that flow. Idempotency keys stop a retried request from creating a duplicate job. Each device receives one welcome grant, which limits abuse of free allowances. Daily ceilings and concurrent-job limits cap how much a single account can consume in a short period. Ilin presents these as the mechanisms that make paid generation behave predictably under retries and failures, which a single API call in a demo never has to handle.
The bug that showed a paid balance to unlimited users
On September 19, 2026, a rendering bug displayed a paid token balance to users who had unlimited access. The cause was a second render that overwrote the unlimited status. Ilin says no charge occurred, but the screen contradicted what the product promised its users.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
The fix was a UI priority rule that decides which status wins when a screen renders twice, plus regression tests that deliberately render twice to catch the same ordering problem. His summary of the lesson is the sentence that gives the article its approach: “The money logic belongs in one place, and the screens only display it.”
Provider policy: test the exact use case first
Dantiva’s photo templates depended on recognizable people. Ilin reports that Veo blocked image-to-video generation with recognizable people in his workflow. He says the filter could not be switched off and that no allowlist was available to him. He continued to use Google models for text-to-video and image generation.
This is a report about his use case at the time of publication, not a general statement about every Veo capability or policy context. His recommendation is direct: “I would test the provider’s content policy on the exact use case before building a whole template catalog around it.” He says he would now make that test before building the catalog, not after.
The alternatives he was evaluating
For photo animation, Ilin was evaluating three other providers. None was connected when he published. His article notes that providers differ on price per second, payment path, and rules affecting Russian users and cards.
| Option | Access route named in the article | Status at publication | Price per second | Payment and rules for Russian users and cards |
|---|---|---|---|---|
| Kling | Through fal.ai | Under evaluation, not connected | Not stated in the article | Not itemized in the article; the article notes providers differ |
| Runway Gen-4 Turbo | Direct | Under evaluation, not connected | Not stated in the article | Not itemized in the article; the article notes providers differ |
| Seedance | Through BytePlus | Under evaluation, not connected | Not stated in the article | Not itemized in the article; the article notes providers differ |
If you are making a similar comparison, the article supports only a few axes: whether the provider accepts your specific personal-photo use case, any consent or recording requirements, price per second, whether payment works for your geography and cards, and the provider’s applicable rules. Ilin does not declare a winner, and his article does not establish that any of these providers accepts the use case today.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Pricing and unit economics
Subscriptions launched on September 18, 2026. The figures below are company-reported prices from Ilin’s article.
| Plan or top-up | Price | Tokens | Clips at 140 tokens each* |
|---|---|---|---|
| Start subscription | 199 rubles | 1,400 | 10 |
| Author subscription | 499 rubles | 4,200 | 30 |
| Pro subscription | 999 rubles | 10,080 | 72 |
| Small token top-up | From 99 rubles | Not stated in the article | Not stated in the article |
*The clip counts are arithmetic from the article’s token figures, not numbers Ilin reports. He states that one eight-second Fast video with sound costs 140 tokens.
Ilin also reports three cost figures. An eight-second clip cost roughly 80 rubles to produce at Google’s list price. The catalog was normalized to around a 30% gross margin at real provider prices. And generation was subsidized by a Google Cloud grant, which he expected to end in November 2026 and said he would use to evaluate the economics afterward.
Recommended Free Tools
Those figures do not let a reader work out margin independently. The article does not itemize realized utilization, how the top-up pricing maps to ruble revenue per token, or how much of the cost base the grant covered. Because the grant was scheduled to end in November 2026, the reported margin describes a subsidized period as much as a steady state. Ilin presents the economics as a hypothesis, and this article treats them the same way.
What the market evidence shows so far
Ilin describes the intended audience as Russian-speaking people who want videos of themselves, such as birthday greetings, social trend clips and avatars. The advantage he says he could substantiate is ruble payment through Telegram with a receipt. He calls the broader demand segment a hypothesis.
At publication, demand was unconfirmed. A later first-customer update reports that one customer completed the purchase path, including a monthly subscription in the Mini App. That update also states that it does not establish repeat retention, a stable acquisition channel or product-market fit. One completed purchase is a useful test of the payment path. It is not evidence that the original uncertainty has been resolved.
A name that changed twice
The customer-facing name moved from Project Aurora to Synora and then to Dantiva. Ilin kept the internal service and environment identifiers as Synora. His reasoning was that renaming them would have meant a risky migration with no benefit to customers. This was a choice for this system, not a general rule against renaming products.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →What Ilin would do differently
- Check the provider’s content rules against the exact intended workflow before building the template catalog.
- Write the ledger before the screens, so that every interface reads from a settled money model.
- Choose the final product name sooner.
- Ship subscription pricing earlier to learn about demand sooner.
These are his retrospective recommendations, drawn from the build he describes. One founder’s account shows which problems surfaced first in one product, not how every AI product will behave. Its value is in the order the problems appeared: the model call worked early, and the trouble came from identity, money, policy and a market that had not yet proved itself.
Quick Recap
“
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
