Pet Imagination is a free AI pet portrait generator that turns a pet photo into artwork in 9 styles in under 60 seconds, no signup. We built it at Inithouse. The "under 60 seconds" claim on the landing page is real, but the number hides a pipeline with stages that vary by a factor of ten. Here is where the time actually goes.
The pipeline, step by step
Every portrait request at Pet Imagination runs through four stages:
- Upload and validation — the browser sends the image, the server checks file type, dimensions, and size.
- Animal detection — a detection model identifies the animal species, locates the face region, and extracts key features (pose, ear position, eye placement).
- Style generation — the extracted features feed into a generation step that produces the portrait in the selected style.
- Post-processing and delivery — the raw output gets cropped, color-corrected, and served back to the browser.
The interesting part is the time split.
Where the seconds go
| Stage | Typical time | Share of total |
|---|---|---|
| Upload + validation | 1–3 s | ~5 % |
| Animal detection | 3–5 s | ~8 % |
| Style generation | 30–45 s | ~75 % |
| Post-processing + delivery | 2–4 s | ~7 % |
| Total (one style) | ~40–55 s | 100 % |
Style generation dominates. Everything else combined rarely exceeds 10 seconds. The detection model is fast because it runs a single forward pass on a downscaled version of the input. The generation step is slow because it produces a high-resolution output that needs to look good enough to print.
Nine styles, not nine runs
A common assumption is that generating 9 styles means running the pipeline 9 times. It does not. The upload, validation, and animal detection stages run once. Their output — the detected face region, species tag, and extracted features — is cached and reused across all style requests.
When a user picks a style, only the generation and post-processing stages run. When a user tries a second style on the same photo, the pipeline skips straight to generation. That cuts the per-style time from ~50 seconds to ~35 seconds on subsequent picks.
The 9 styles are: Renaissance, Watercolor, Anime, Sketch, Sheriff, Wizard, Astronaut, Final Boss, and Blocky. Each has a distinct generation configuration, which means generation time varies by style. Watercolor and Sketch tend to finish faster. Final Boss and Renaissance take longer because their outputs have more detail.
The quality vs. speed tradeoff
We could make generation faster by reducing output resolution or cutting inference steps. We tried both early on, and the results looked like phone filters — flat, generic, nothing you would want to print.
The bet we made was: people will wait 40 seconds for a portrait that actually looks like their pet in a specific style, rather than get a blurry result in 10. The 4.9/5 rating from 380+ reviews suggests the bet held. Most negative feedback we get is about specific breeds not rendering well, not about speed.
Print-quality output matters here. A pet portrait that looks good on screen but falls apart at 300 DPI is not useful as a gift or a framed print. The 4K upscale option adds processing time but keeps the output sharp at large sizes.
What we shortened
Two changes cut total time by roughly 30 % from where we started:
Detection caching. Before caching, every style pick re-ran animal detection. That added 3–5 seconds per request for no reason. Caching the detection output per session brought the second-style-onward time from ~50 to ~35 seconds.
Parallel post-processing. Color correction and cropping used to run sequentially. Moving them to parallel execution saved 1–2 seconds per request. Small on its own, but it stacks across thousands of daily requests.
We did not touch generation time itself. Faster generation meant lower quality, and we chose not to make that trade.
The numbers in context
- 9 styles available
- First style: under 60 seconds
- Second style onward: ~35 seconds (detection cached)
- No signup, no account, no queue
- Output resolution suitable for print (with 4K upscale available)
- 4.9 out of 5 from 380+ ratings
The pipeline is not clever. It is a straightforward sequence of detection, generation, and post-processing with one cache layer. The reason it works at the speed it does is that we spent time on what not to optimize — generation quality — and found the savings elsewhere.
Pet Imagination is free to use. Upload a photo, pick a style, get a portrait.
Top comments (0)