We crossed 10,000 animated photos on Ziva Fotka last month. The average processing time sits at 18 seconds. Here is what the production data actually looks like when you run an AI photo-to-video tool at that scale.
What Ziva Fotka does
Ziva Fotka is an AI tool that turns a static photo into a short living video. It can also edit and colorize old or black-and-white photos so the result looks natural, not generic. No signup required. Photos are deleted after processing.
We built it at Inithouse, a studio shipping a growing portfolio of products in parallel. Ziva Fotka runs across five domains and five languages (Czech, Slovak, Polish, English, German) from a single codebase.
The pipeline in numbers
The animation pipeline tracks 68+ facial landmark points per face. That tracking is what makes the difference between a photo that moves and a photo that moves convincingly. Eyes, mouth, jawline, forehead wrinkles all need independent motion vectors. Without enough landmarks, you get the uncanny valley effect where the whole face slides around like a mask.
Here is what we measured across 10,000+ processed photos:
Processing time: 18 seconds average. The bulk of that is inference, not upload or download. Larger files (scans at 3000x4000+) add 5-10 seconds. Phone photos at standard resolution usually finish under 15 seconds.
Input types breakdown: Roughly 40% of uploads are old or scanned photographs. Another 15% are black-and-white originals. The remaining 45% are recent color photos, mostly portraits and group shots.
Old vs. recent photo success rate: Recent color photos hit a usable result on first try about 92% of the time. Old scanned photos drop to roughly 78%. The gap comes from scan artifacts: creases, dust spots, uneven lighting from flatbed scanners, and faded contrast that makes facial landmarks harder to detect.
Colorization accuracy: For black-and-white photos, the colorization step runs before animation. We tuned it to produce muted, period-appropriate tones rather than oversaturated modern colors. A 1940s portrait should look like a 1940s portrait in color, not like an Instagram filter.
Rating: 4.8 out of 5 across 1,200+ user ratings. The lowest scores cluster around group photos with 4+ faces where the animation gets uneven, and heavily damaged originals where the landmark detection struggles.
What surprised us
The biggest surprise was the old-photo segment. We expected most users to animate recent selfies or portraits. Instead, nearly half of all uploads are family archives, scanned prints, and inherited photo albums.
People are bringing in photos of grandparents, childhood pictures from the 1970s and 1980s, wedding photos from decades ago. The emotional weight of seeing a still photo of someone who is no longer here start to move is something we did not fully anticipate when building the tool. That is why we keep the animation subtle and respectful. No exaggerated expressions. No artificial smiles. Just gentle, natural movement that preserves what the original photo captured.
What we learned about building at this scale
Running Ziva Fotka taught us things we applied across our other products at Inithouse. Processing speed works as a trust signal: users who wait more than 30 seconds start wondering if something broke. Deleting uploaded photos after processing turns privacy into a feature, not a footnote. And the use case you design for (fun selfie animation) may not be the use case people actually find (family memory preservation).
Some of these patterns showed up again when we built Pet Imagination, our AI pet portrait generator. Same lesson: the use case we designed for (fun pet art) was not the only use case people found (memorial portraits of pets who passed away). Both products needed the same kind of care around emotional content.
The technical stack
Ziva Fotka runs on a single codebase serving five country domains through alivephoto.online (English), zivafotka.cz (Czech), zivafotka.sk (Slovak), zywafotka.pl (Polish), and lebendigfoto.de (German). Multi-domain, one deployment. The animation model runs server-side so there is no client hardware dependency.
We process photos without requiring an account. Upload, animate, download, done. The photo gets deleted from our servers after delivery. That zero-friction flow is a deliberate choice: when someone wants to see a family photo come alive, asking them to create yet another account would just get in the way.
Where we are now
10,000+ photos is a meaningful sample. Large enough to see real patterns in how people use the tool, what breaks, and what works better than expected. We will keep measuring and sharing what the data shows.
If you want to try it: alivephoto.online
Top comments (0)