Thumbnails that keep their promise.

Generating a hundred thumbnails was never the bottleneck. Knowing which one to ship is.

URIM produces genuinely different options, then screens them the way a viewer actually sees them — before you spend a test slot finding out.

See how it works

Thirty candidates, all plausible. Producing them is the cheap part.One survives the rubric. That judgement is the bottleneck — and it is the part URIM is actually built for.

The one thing to understand

Nobody sees your thumbnail at full size.

They see a small tile on a phone, for about one second, next to your title, competing with eleven other videos. That is the only viewport that matters — and it is why AI thumbnails so often disappoint. Toggle between a composition built for the tile and one built for the poster, then compare them at the size that counts.

What you review — 1280 × 720. At this size the detailed one arguably looks better.

Your video

What viewers get — roughly browse size, in a row of competitors. One of these still reads in a second.

Your biggest lever

The headline isn't baked into the image.

It's a separate layer, so you can rewrite it on any candidate instantly — no re-render. Most tools bake the text in, which quietly costs you the ability to test copy at all. Try it:

Headline

2 of 4 words

Testing copy is nearly free. One good image, four headlines, four real options.
Typos are a two-second fix, not a regeneration.
Other languages are a text swap, not a new project.
Your actual brand typeface, not the model's approximation of it.

How it works

URIM does the volume. Every judgement call stays with you — and the last step feeds back into the third, so it gets tuned to your channel rather than to general advice.

01Upload

02Review the brief

03Pick directions

04Generate

05Review scores

06Edit the text

07Ship three

08Learn

what actually won feeds the next brief

Your call — 6 of 8URIM works alone

01
Your call
Upload
Your video and a working title. No brief to write.
02
Your call
Review the brief
The payoff, the peak moment, the curiosity gap, and what the video actually delivers. Errors are cheapest to fix here.
03
Your call
Pick directions
Concepts arrive as words, before anything is drawn. Structurally different from each other — not one idea in six colours.
04
URIM
Generate
Artwork for the directions you kept. A saved face reference keeps recurring people looking like themselves.
05
Your call
Review scores
Each candidate is scored at browse size on clarity, legibility, contrast, emotional read, and curiosity.
06
Your call
Edit the text
The headline is a separate layer. Rewrite it on any candidate instantly, with no re-render.
07
Your call
Ship three
Into YouTube Test & Compare, which picks the winner by watch time rather than clicks.
08
URIM
Learn
Results feed back, so suggestions get tuned to your channel rather than to general best practice.

The guardrail

Sometimes it discards the best-looking option.

If you only optimise for clicks, you get clickbait. That isn't a moral argument — it's a mechanical one. YouTube watches what happens after the click. A thumbnail that wins clicks and loses viewers gets suppressed: you get a spike, then a colder algorithm than you started with.

OverpromisedKept its promiseIllustrative — the shape, not measured data.

So every candidate is checked against what your video actually delivers, and the ones that overpromise are cut — however well they score elsewhere. That's the guardrail working. It's protecting the channel, not the thumbnail.

What it doesn't do

Where our claims stop is as useful to know as where they start.

URIM's job

Produce the volume
Score legibility at browse size
Composite the text layer
Cut the overpromises

Yours

Whether the video is any good
Who your audience is
The final pick

Nobody's to promise

A guaranteed lift

It won't fix a weak video

A thumbnail sells the click; the video keeps the viewer. High CTR on a video people leave is worse than no click.

It won't replace your judgement

Scoring catches technical problems — illegible, low contrast, unreadable at size. It does not know your audience.

It isn't instant

Generation runs in minutes. Plan it as a step in your upload process, not something you do while the upload bar fills.

It doesn't guarantee a lift

Testing three good options beats shipping one guess. That is the honest claim.

Want it when it's ready?

URIM is in development. Leave an address and we'll get in touch when there's a version worth your time — and tell you plainly what it can and can't do yet.