What this place actually is
ShinobiTools is one person’s set of tools that share a pass and a set of machines. There
is no company floor behind it. The main machine sits in a room with two
consumer graphics cards in it, and a few more computers that I also run myself take the other heavy
jobs. When a tool here says it removed a background, one of those machines removed it.
That matters more than it sounds. I pay for the hardware once and the electricity monthly,
so what I can afford to give away depends on my own costs and not on anyone else’s
pricing. The trade-off is fair and worth knowing: the cards are a shared resource, so at a
busy moment a heavy job waits.
The plain tools (cleaning pasted text, counting it, converting an image, stripping
metadata) are free without a limit and without an account. Anything that costs real
machine time has a daily number instead: the GPU jobs, but also the AI rewrite and the
image-to-vector conversion, which are capped per day even though they look light. If you need
more than that on a given day, a pass is €5 for 30 days or €3 for 24 hours,
and the same pass works on every tool in the family rather than only the one you bought it on.
I do not sell your files and I do not keep them for later: uploads and results are deleted automatically. The tools work without an account; only
buying a pass asks for an email address, so the receipt and the pass have somewhere to go.
Where these tools stop
A tool that lists only what it can do is advertising. Here is the other half, because it
saves you the upload.
Background removal is good at a person, a product or an animal against a reasonably separate
background. It gets hair against a busy background wrong, and it cannot invent an edge that
the photo never captured. Upscaling adds plausible detail, not real detail: a license plate
that is four pixels wide stays unreadable, however sharp it looks afterwards. Photo restoration
can take out scratches and sharpen soft detail, but it does not correct fading, and a face that has lost its features is gone.
Stem separation needs a mix. A finished studio mix gives the splitter the most to work with;
a phone recording of a live room has the room mixed into everything, so expect some of it to
come along. Transcription depends on how clear the recording is: when two people talk over each
other the words are hard to tell apart, so read the result through and correct it. OCR wants a flat, straight scan at a sensible resolution; a
photo of a curved page taken at an angle is where it gives up.
What running twelve sites side by side has taught me
Twelve sites sharing one set of machines teach you things a single-purpose site never sees. The clearest one: a limit that looks arbitrary from outside is almost always a queue problem from inside. A GPU can only hold one big job at a time, so a video background removal and a photo restoration compete for the same card; the free-tier caps on file size and length keep that queue moving for everyone, and they are also what a pass raises.
The tools that run in your browser (resizing an image, converting between common image formats such as HEIC, JPG, PNG and WebP, the Discord size tool) don't have that problem at all, because nothing leaves your device. Video, audio, e-book, camera RAW and image-to-SVG conversions are the exception: they run on the server, with caps on the free tier. That's a deliberate split, not a coincidence, and it's why some jobs feel instant while others show a queue position. The other pattern: people usually arrive already knowing the file type they have and not the tool name for it, "this HEIC won't open" rather than "I need a converter", which is the whole reason the tool chooser and the grid above are organized by task, not by tool name.