Sketch pad
OptionalDraw the layout you want — subject left, text right, focal zones. The AI respects your sketch as a composition blueprint.
Paste any YouTube channel — yours, a competitor's, one you admire. We extract its visual DNA — palette, typography, composition — and bake it into every render. No more "looks AI-made".
Pick a style source. Type your title. Generate. Every tool below is yours to enable or skip — none of them are required.
Paste any YouTube channel — yours, a competitor's, or one you admire. We read its visual DNA from the last 20–50 thumbnails: palette, typography, composition, character treatment. Every render inherits the look.
Got one specific thumbnail you love? Paste its YouTube URL or drop the image. We extract its EXACT style — colors, layout, lighting, character render — and apply it to your new title.
Every tool below is opt-in. Use none and the render still works. Use all and the render gets dramatically better.
Draw the layout you want — subject left, text right, focal zones. The AI respects your sketch as a composition blueprint.
Drop up to 4 mood / subject / element images. Each one feeds the renderer as inspiration alongside your channel style source.
Title says "Elon Musk"? We auto-fetch his Wikipedia portrait and ground the render against it. No more plastic-faced AI hallucinations.
Three dials: Copycat hugs the reference style tightly. Be Creative invents. Mixed splits variants across both.
1, 2, 4, or 8 different renders in parallel from the same title. Each one's a different creative take — pick the one that hits. Costs scale linearly with token spend.
Add your typography in Photoshop later? Flip this on and the renderer leaves a clean canvas — no on-image text at all.
Don't love a variant? Three ways to iterate without burning a fresh render from scratch.
Describe a fix in plain English: "darker palette, push the subject right, remove the small text." The next render obeys exactly.
Circle any region. Add a note: "make it red", "replace with a tighter zoom". The next render respects that exact pixel range.
Same subject, different channel's DNA. Take your rendered thumbnail, pick another channel — get the same scene in their visual language.
Most AI thumbnail tools skip six of them. That's why ours doesn't look AI-made.
Our vision pipeline reads the reference channel's last 20 to 50 thumbnails and extracts a style fingerprint. Every render inherits it.
Title mentions Elon Musk? We pull his Wikipedia portrait and ground the render against it. No more plastic faces.
Your title gets a 0–10 pairing score against a suggested on-image hook, computed from the reference channel's text DNA.
Draw a layout in-browser. Subject placement, composition zones — the AI respects your sketch as a composition blueprint.
Drop up to 4 mood / subject / element images. Each one feeds the render engine as inspiration alongside the channel DNA.
Circle any region, add a note ("rotate this", "darker"), and the next render respects exactly that pixel range.
Every refine, PinPoint edit, and restyle is auto-saved. Switch between versions in one click — nothing is ever lost.
Keep the scene, change the visual treatment. Restyle your render in a different channel's DNA without losing the composition.
A curated set of thumbnails from creators known for thumbnail craft. One click to reference any of them in your next render.
A fair question. Here's what changes once you actually ship a video.
Paste any YouTube channel URL. We pull its last 20 to 50 published thumbnails and run them through our vision pipeline to extract a visual fingerprint: palette, typography, character treatment, composition patterns. That fingerprint is baked into every new render, so the output reads as "the same channel's next thumbnail" instead of generic AI art.
The second generation path. Got one specific thumbnail you love? Paste its YouTube URL or drop the image directly. We extract that exact thumbnail's style (composition, colors, lighting, character render) and apply it to your new title. Useful when you want to clone a single hit's formula rather than a whole channel's aesthetic. Pro and Studio only.
Yes, perpetual. 30 tokens per month (one render of every paid feature you'd see on the higher tiers), with a small watermark on output. Every render still runs on our premium image engine, the same one the paid plans use. No credit card. The free tier is a real product surface, not a 7-day trial.
360 tokens per month (about 12 thumbnails), the watermark removed, and 2 variants per render so you actually have something to A/B test. It's our cheapest paid plan and the right fit if you ship a handful of thumbnails a month and care about clean, watermark-free output.
Yes. Unused subscription tokens carry over to the next month. Every billing cycle adds your plan's grant on top of whatever's left from the previous one, so you never lose what you didn't spend. Add-on token packs also never expire.
30 to 60 seconds per single-variant render. Multi-variant runs (1, 2, 4, or 8 in parallel) take the same wall-clock time because they generate concurrently. The very first time you reference a new channel, add roughly 10 seconds for the DNA scrape. After that the channel is cached.
Our character detector scans the title for named entities (people, brands, fictional characters). When it finds one, we pull a Wikipedia portrait and feed it to the render engine as a likeness reference. The render respects the real person's actual features instead of inventing a plastic-faced approximation. Toggle the detector off if you want a stylized or anonymous look.
Three ways. (1) Refine: describe a change in plain English ("darker palette", "remove the small text"). (2) PinPoint: circle a region and annotate it ("make this red", "rotate this 15°"). (3) Restyle: swap the whole thumbnail to a different channel's DNA in one click. Every version is auto-saved, and you can switch between them anytime. Refine and PinPoint are on every plan. Restyle is Pro and Studio.
Canva AI ships templated thumbnails: no channel memory, no entity grounding, plastic-skin faces. Pikzels is the closest competitor and shares the channel-DNA idea, but CopyThumb adds Clone-a-thumbnail mode, Wikipedia-portrait grounding for named people, PinPoint annotation editing, and Restyle to swap a render's DNA in one click. ThumbnailTest focuses on A/B testing existing thumbs, not generating new ones, so it's complementary rather than competitive.
Especially well. Faceless thumbnails are composition + text + visual metaphor, all of which the pipeline is tuned for. Entity grounding also gives faceless channels access to historical, scientific, and Wikipedia imagery without needing a person on camera.
Yes to both. Cancel from your account billing tab in two clicks. Your plan stays active through the end of the period you already paid for, then reverts to Free. 14-day money-back guarantee on your first paid charge, no questions asked. Payments are processed by Paddle (Merchant of Record), who handles VAT and sales tax globally.
No. Saved reference channels and generated thumbnails live in your account only. Our image providers' paid-tier terms explicitly forbid training on customer inputs, and we never opt in. Your prompts and renders are never used to train models, ours or anyone else's.
Start free, no card. DNA extraction on any channel takes 30 seconds.
