Sprites
Making a visual novel and staring at 650 AI-generated sprites? Queue them up, send the link to the art department, and let the verdicts flow back to your agent to sharpen the prompt.
Turn a batch of generated outputs into human feedback an agent can use.
Send your data to a queue. Share that queue with friends or colleagues for review. Get quick feedback from real humans.
$ npx -y trainything
on npm soon
create_queue title="Batch 07 sprites"
intro="Reference: Mira, red bob haircut, green scarf.
Reject anything off-model."
feedback=on_reject
add_cards paths=./out/batch-07/*.png → 40 cards
share_queue → https://…/r/batch-07 · sent to a phone
… three minutes later …
get_results format=summary
28 kept, 12 rejected. Notes cluster on:
hair drifting to orange (7) · scarf missing (4) · extra fingers (1)
→ the agent pins "red bob" and "green scarf, always visible" and re-runsHow it works
1 · THE AGENT CREATES
A queue of things to judge
From Claude Code, Codex or the shell. One question, one gesture per card.
create_queue title="Batch 07 sprites"
intro="Reject anything off-model."2 · A PERSON JUDGES
One card at a time, on a phone
A link. No account, no training. Tap, tap, tap.
https://…/r/batch-07
Accept · Reject · Skip
"why?" only on a reject3 · THE AGENT READS BACK
Verdicts, notes, consensus
A file it can act on: JSON, labels, or a summary to read.
get_results format=summary
28 kept · 12 rejected
notes: hair (7) scarf (4)Works out of the box: ask your agent, or run it straight from the CLI. Send a link to whoever should review, and they open it on a phone or a desktop.
What it can do
single
Says yes or no.
a label
pair
Picks this or that.
a preference
choose
Picks one of N.
a class
select
Picks any of N.
tags
match
Pairs left with right.
an alignment
Cards can be text, images, audio or video. The gesture you pick is the shape of the data you get back.
Who uses it
Dialogue
A whole novel's worth of lines and no idea who said what? Send them to your cast and have them sort each line to its speaker, quickly, from their phones.
Free and paid
Free to run yourself. We sell running it for you.
Community
$0
The whole product on your machine, your LAN, your tunnel or your own server. Every mode, every format, the dashboard. The link lives only while your laptop does.
npx -y trainything on npm soon
Cloud
from $15 a month
Your own instance on our domain. The link lives on when your laptop closes. Media on a CDN. Teammates with their own keys from $49. An address your agent can reach from anywhere, including claude.ai.
Coming. Join the list above.
Source-available and free to run yourself. Licence text ships with the package.