The CapCut AI thumbnail maker is the obvious first try if you already edit in CapCut: you describe a picture or upload a reference image, a model draws a base image, and the words go in by prompt or as an overlay in CapCut's built-in design editor. A reference-based generator such as WThumb runs the other way: you paste a thumbnail you admire, type the words, and the words land in the picture.
Neither needs a label. As of September 2026, YouTube's Disclosing use of GenAI content page lists production assistance, like using generative AI tools to create or improve a video outline, script, thumbnail, title, or infographic, among the uses that do not require disclosure.
This post reads CapCut's page for the inputs, the face, the words, the export advice and the Shorts gap, and sets each beside the same step in WThumb.
What CapCut's AI thumbnail maker takes in
This section lists the inputs CapCut's page names. The AI Thumbnail Maker page, read in September 2026, offers two ways in: a text prompt, with the examples excited youtuber pointing at text that says new and shocked face with red arrow, or an uploaded image. It names Seedream 5.0 and Nano Banana Pro among its models; the second is the model that Nano Banana YouTube thumbnails looks at on its own. After the first image you can, in the page's words, keep chatting with the AI: make the background brighter, or add a yellow arrow. The page calls the tool a free AI YouTube thumbnail maker.
A description is a blank page: the model invents the subject, the framing and the tone from one sentence, and shocked face with red arrow can come back a hundred ways. A reference is a brief: you hand over a picture that already has the energy you want and ask for that idea with your subject in it. In WThumb the reference is a pasted YouTube link (only that video's public thumbnail is fetched) or a PNG, JPEG or WebP upload, and it is a brief, not a template: the idea carries over, the design is new. CapCut's upload box takes a reference image too; the difference is what happens next.
Your face, and the AI face generator
This section is about whose face ends up in the thumbnail; the two tools answer it differently. CapCut's page answers directly: yes, upload a photo of yourself or one generated using the AI face generator, then describe the expression and background you want, and the AI will place your face into a new scene. So a CapCut run can carry your real face, or a face that belongs to nobody.
A generated face is fine as long as it stays nobody's. As of September 2026, YouTube's Impersonation policy page lists using AI to make it look like a famous person is willfully participating in your video or endorsing a product, and using someone's AI-generated likeness or voice to falsely imply they own or authorize your channel's content, among the things it does not allow. A generator that drifts toward a known streamer's look, or a prompt that asks for one, is what that page describes. A face that resembles nobody is the only safe version.
In WThumb the choice has two options and the second one is empty. Add your photo (up to 8 MiB) and you are placed in the design. Skip it and the run is faceless: the person in the reference is removed and the space is filled in the same style. Nobody else's face is ever put in, and there is no face generator to drift. For a review titled WEIRDEST GADGET, that is you holding the gadget, or the gadget alone and large.
Where the words come from
This section follows the words, the largest difference between the two. CapCut's page describes two routes. Its opening copy says the tool handles composition, lighting, and text placement, and one of its prompt examples is excited youtuber pointing at text that says new, so a prompt can ask for words in the picture. Its FAQ describes the other route: after generating the base image, use CapCut's built-in design editor to overlay text, arrows, circles, or emojis, a two-step process the page says gives you full control over the final look. On the second route the text sits on a picture that was not drawn around it, so you pick a corner and a color that reads against whatever the model put there.
In WThumb the words are part of the brief. You type up to three lines (WORTH IT? on one line does more work than IS IT WORTH IT? on two) and one line of instructions in plain words, and the words land in the picture: sized, placed and colored as part of the design. There is no canvas, no layers and no design step after the run. If the words came out too small, the pencil on the finished thumbnail takes one sentence, make the words larger and yellow, and makes a new version.
Which you prefer is about how you work, and do you need a thumbnail editor or a generator goes through it. If you like placing text by hand, the overlay step suits you. If you want to type the words once and get a finished file, the brief route is one step instead of two.
A brief to paste
- Words. WORTH IT? on one line.
- Instructions. Me holding the gadget at chest height, surprised, the words large on the empty side, one warm accent color.
Export: 1280 by 720, PNG, and the 2 MB line
This section checks CapCut's export advice against YouTube's own numbers; they differ in one place. CapCut's page tells you that YouTube recommends 1280x720 pixels with a 16:9 aspect ratio, that the tool lets you set custom dimensions, to always export as PNG for best quality, and to keep file size under 2MB. As of September 2026, YouTube's Add custom thumbnails page says a thumbnail should have a resolution of 3840 x 2160 pixels for videos and 2160 x 3840 for Shorts, with a minimum width of 640 pixels for videos, and be uploaded in image formats such as JPG or PNG. On file size, the same page says Mobile: 2 MB for video thumbnails or 10 MB for podcasts. Desktop: 50MB for video, Shorts, and podcast thumbnails.
So CapCut's 1280 by 720 clears the 640 pixel minimum and sits well below the 3840 by 2160 YouTube names; if the custom dimensions go higher, use them. And 2 MB is the mobile ceiling, not the desktop one: a PNG that comes in at 3 MB uploads from a computer and is over the limit in the YouTube app. If you finish on your phone (make a YouTube thumbnail on your phone walks that path), the 2 MB advice is right.
A WThumb run returns a Full HD+ JPEG, sharper than 1080p: above the 640 pixel minimum, below the 3840 by 2160 recommendation, and in a format the same page names. PNG against JPEG matters less than it sounds, because YouTube re-encodes every upload anyway. The thing to protect is the source: a large, sharp image with big words, exported at full size and never screenshotted.
Shorts: what CapCut's page does not say
This section covers the vertical case, because much CapCut editing is vertical and its thumbnail page is not. The AI Thumbnail Maker page's export advice is 16:9 only. It adds that the tool lets you set custom dimensions, but has no line about 9:16, Shorts or a vertical output. That may be the page rather than the tool, so check whether a custom 9:16 works before you plan a Shorts cover on it.
As of September 2026, YouTube's Making thumbnails easier post, dated July 2026, says YouTube Partner Program creators can start uploading custom thumbnails for Shorts, that it will expand this to more creators over time, and that desktop now offers three suggested thumbnail frames for Shorts. Read in September 2026, YouTube's Add custom thumbnails page gives the Shorts numbers, 2160 x 3840 and, for Shorts, try to upload a thumbnail with an aspect ratio of 9:16, and adds that custom thumbnails for Shorts are currently only available to add in YouTube Studio on a computer. A 16:9 file cropped to 9:16 loses both sides, which is usually where the words were.
In WThumb the vertical version is a tick, not a crop. One submission can have 16:9 and 9:16 ticked together, so one brief produces a YouTube thumbnail and a Shorts thumbnail, each at its own shape. If you are not in the Partner Program yet, the 9:16 file waits; the 16:9 one uploads today. Write the words with the tall frame in mind: I TRIED IT stacks on two short lines in either shape.
Which one, and when
This section is the decision, and it turns on where you start. Four starting points cover most creators. Whichever you pick, the seven decisions in how to make a thumbnail for YouTube still apply.
- Already in CapCut, want overlays. CapCut. Base image, chat refinements and overlay step are all its own; text, arrows, circles, or emojis go on in its built-in design editor.
- Have a reference and the words. A reference-based run. Paste the link, type the words, and the words are in the picture without a second step.
- Need 16:9 and 9:16 from one brief. In WThumb both are ticks on one submission. Check whether CapCut's custom dimensions give you a usable 9:16; its page does not mention one.
- Want to retype the words next week. On CapCut's two-step route the text is an overlay in its design editor, so new words are an edit. In WThumb the words are in the picture, so new words mean a new run.
Both are free to start. CapCut's page calls the tool a free AI YouTube thumbnail maker and states no prices, credits or watermark behavior, so check CapCut's own pricing page for what free includes. In WThumb, new accounts get two free credits (1 credit = 1 thumbnail), no card needed, and the paid plans are on the pricing page. For the wider field, the best AI thumbnail makers sorted by how they work puts both approaches beside the rest. Five of the dedicated makers are set side by side on price and size in our AI thumbnail maker comparison.
How to do this in WThumb
This section is the whole run, for the creator with a CapCut project open and a thumbnail they admire in another tab.
- Paste the reference. Paste the YouTube link (WThumb fetches only that video's public thumbnail) or upload a PNG, JPEG or WebP. It is a brief, not a template.
- Type the words. Up to three lines: I TRIED IT on one, WORTH IT? on the next.
- Write one line of instructions. Plain words: me holding the gadget at chest height, surprised, the words large on the empty side. Attach up to four extra images and point at them as @1 to @4.
- Add your photo, or leave it out. With a photo you are placed in the design. Without one the run is faceless, with the person in the reference removed.
- Tick the formats. 16:9, 9:16 or both, and one or two designs from the same brief.
- Wait about thirty seconds. Your words are checked by a moderation service before anything is generated; a refused request is never generated and costs nothing.
The @1 trick for the product
If the video is about one product, attach its photo as @1 and write @1 in my hands in the instruction line. Extra images are briefs too, not templates: the product is worked into the design rather than pasted on top.
Revise it, or run it fresh
Press the pencil on the finished thumbnail and describe one change in plain words, the words bigger and yellow, and a new version is made. Each revision costs one credit, and the new version can be revised again. If the whole idea is wrong, a fresh run from a better brief beats a string of revisions.
One tool asks you to describe a picture and then finish it. The other asks you to point at one and type the words.
