Tell us the job. Then go do something else.
You spend about ninety seconds describing what you want. After that you are not needed. Pick a job to see exactly what happens, step by step.
What are you trying to get done?
JOB 01 · Images
Clean up a pile of product photos
Mismatched backgrounds, inconsistent crops, some shot on a phone three years ago. You want them all looking like they belong to the same store.
- Say what you want“White background, same crop, sharp enough for the product page.” Plain words are fine.
- Point us at the photosAll of them at once. Nothing needs renaming first.
- Confirm the costYou see the price before anything starts. Approve it and the work begins.
- Reads every photo and sorts out which ones need rescuing
- Lifts each product off its background
- Crops and places everything on the same clean white
- Sharpens the older, softer shots up to size
- Puts the finished set together where you can get at it
What you will have when it is done
- The full set, renamed and numbered, ready to upload
- A preview grid, so you can spot-check before downloading anything
- The handful it could not save, flagged separately with a reason
What is actually running
Background removal with BiRefNet, upscaling with Real-ESRGAN, and Qwen-Image-Edit for shots that need their lighting repaired. These are editing models rather than generators, which is why your actual product is preserved instead of redrawn.
JOB 02 · Documents
Make a stack of documents searchable
Hundreds of PDFs. Some clean, some scans of scans, most with tables that matter more than the paragraphs around them.
- Say what you need out of them“Pull the date, the total and the client name from each one, and let me search across all of them.”
- Point us at the filesScans, exports, photos of paper. Mixed formats are fine.
- Confirm the costApprove the price and step away. The files stay on the machine you started.
- Reads every page, including the crooked scans
- Keeps tables as tables instead of scrambling them into text
- Pulls the fields you asked for into one consistent sheet
- Builds a search that works across the whole stack
- Writes a summary of what the documents have in common
What you will have when it is done
- A spreadsheet with one row per document and the fields you named
- A search box that answers questions in plain English across all of them
- A written summary of the patterns across the whole set
What is actually running
Page reading with PaddleOCR-VL, charts and figures handled by Qwen3-VL, field extraction and the summary by Qwen3, and search built on BGE-M3 embeddings. All of it runs on your session and is destroyed with it.
JOB 03 · Video
Make short videos for social
You need a lot of clips, they need to look like your actual product, and you do not have a studio.
- Describe the clip and add a product photoThe photo is what keeps the product accurate. Text alone lets it drift.
- Say how many and what shape“Five per product, vertical, about eight seconds.”
- Confirm the costThen step away. There is one short check-in later.
- Makes a quick rough version of every clip
- Pauses and shows you the drafts — you keep the good ones, about 2 minutes of your time
- Re-makes only the keepers at full quality
- Smooths the motion so nothing stutters
- Exports everything sized for each platform
What you will have when it is done
- Finished vertical clips, one file per cut, ready to post
- The rejected drafts, kept in case you change your mind
- A contact sheet of first frames, so you can find things fast
What is actually running
Drafts by LTX-Video because it is fast and cheap, finals by Wan 2.2 image-to-video starting from your real photo, and RIFE for frame smoothing. Drafting first is why this costs a fraction of doing every clip at full quality.
JOB 04 · Audio
Turn recordings into usable text
Interviews, episodes, calls, webinars. Hours of it, none of it searchable, and no idea what is in there.
- Say what you want out of it“Transcripts with who said what, chapters, and the bits worth clipping.”
- Point us at the audioAny length. Nothing needs splitting up first.
- Confirm the costApprove and go. This one is usually done before you remember to check.
- Transcribes everything far faster than real time
- Works out how many voices there are and labels who is speaking
- Lines the words up to the second, so timestamps are exact
- Finds the natural chapter breaks and the quotable moments
- Drafts notes and a summary for each recording
What you will have when it is done
- A transcript per recording, speakers named and timestamps throughout
- Chapter markers you can paste straight into your player
- A clip list — the strongest moments, with exact start and end times
What is actually running
Transcription by faster-whisper on the large-v3 model, speaker separation by pyannote, and Qwen3 for chapters, clips and notes. Speaker labelling is the piece chat tools do not do at all.
Start a job when rentals open.
Rentals are opening to a small group at a time, and the first 100 accounts keep the rate they join at for twelve months. Leave an email and we will tell you when your turn comes. One message, no newsletter.
Questions people ask
What happens in the ninety seconds?
You say what you want in plain words, point us at the files, and confirm the price before anything starts. There is no configuration screen, no hardware to choose and nothing to install. That is the whole of your part on most jobs.
Do I get the finished files emailed to me?
No. You get a link to the work already running, and a person available the first time to make sure it does what you wanted. Drop-off-and-collect batch processing is not something we do yet, and we would rather say so than imply otherwise.
What if my job is not one of these four?
Describe it in your own words in the box below the cards. These four are common shapes rather than a menu, and the matching works the same way whether your job resembles one of them or not.
How do I know what it will cost before it runs?
The price is shown and confirmed on screen before anything starts, and you set a spend cap that nothing is allowed to exceed. A live meter shows what has been spent and what is left for the whole time a session is open.
Are the times on this page real?
They are illustrative, modelled on jobs of that size rather than measured on our own infrastructure. They will move with file sizes and with how much rescue work your source material needs, which is why the price is confirmed against your actual files rather than against these.
Times and costs are illustrative, modelled on a job of this size rather than measured on our own runs. Every session shows the price and a spend cap before anything starts.
Not sure this is your kind of job? What AI can do that a chat window cannot is the longer argument, with four of these worked end to end.