Gaming video thumbnail
Big face, strong expression, saturated background taken from the game. Generate both variants with two different emotions rather than two backgrounds: the face decides the click, the background only sets it off.
AI YouTube thumbnail maker
Create clickable YouTube thumbnails with your own face: one creator card, your video title, and two 16:9 thumbnails generated side by side so you can compare before publishing.
Interactive demo
Real workflow, real results. Pan and zoom freely.
Use this workflowThe workflow above is real: the creator card feeds two competing thumbnails, the title is wired into both prompts. Pan around, zoom in, open the nodes.
YouTube thumbnails with AI: your face, two A/B versions for 10 cents
At a glance
- What the workflow produces
- 2 images in 16:9
- Canvas structure
- 4 nodes wired by 2 connections
- Models wired in
- Seedream 5 Pro
- Cost of one full run
- about 10 credits, roughly €0.10
A good thumbnail comes down to three ingredients: your face with a readable emotion, short high-contrast text, and a background that pops. This workflow assembles them: a Reference card holds your photos from several angles, a Text node carries the video title, and two Image nodes generate two art directions in parallel.
That is the real advantage of a canvas over a plain generator: variants A and B share the same references and the same title, but each keeps its own prompt. You change one background without touching the other, regenerate only the one that hesitates, and your face stays yours from one generation to the next thanks to the Reference card.
Model-wise, Seedream 5 Pro excels at text inside the image and face fidelity; Nano Banana is the fast alternative for exploring. The guide writing a prompt that works covers describing an expression, and the public calculator shows each model's exact cost before you even sign up.
How it works
- 1
Fill in the creator card
Add 2 or 3 photos of your face from different angles and describe your style. The card is @-mentioned in the prompts: the model receives your real photos as references.
- 2
Write the video title
The Text node holds your title. It is wired into both thumbnails: change it once and both variants update.
- 3
Generate variants A and B
Each Image node has its own art direction: shocked expression on orange, confident smile on blue. Keep, tweak or regenerate each variant independently.
- 4
Download in 16:9
Thumbnails come out in YouTube's format. Check readability at small size: a thumbnail is judged at the size it appears in the suggestions feed.
Variations and use cases
The same canvas, tuned for different needs: each variation is two or three lines of prompt away.
Podcast or interview thumbnail
Two faces, a short title, a dark background. Use two Reference cards in the same prompt, and plan a wide shot: models handle two tightly packed heads badly.
Vertical Shorts thumbnail
Switch the Image nodes to 9:16 and lift the subject: on a Short, the lower half is covered by the title and the buttons. Nothing else on the canvas changes.
Thumbnail without Photoshop or cut-outs
There is no cut-out here: the face comes from the Reference card and the background is generated around it. That replaces the hour spent masking a portrait and hunting for a royalty-free background.
A/B test before publishing
The two Image nodes share the title and the references, each keeps its own prompt. You get two comparable thumbnails, to publish one after the other on the same video and measure click-through.
Common mistakes
A long title inside the image
Three or four words maximum: beyond that, models warp the letters and the text is unreadable at thumbnail size. The full title lives on YouTube, not in the image.
Judging the thumbnail at full size
A thumbnail is read two centimetres wide. Shrink the render to that size before choosing: half the compositions that look great full screen do not survive the real format.
A single photo in the creator card
With one angle, the face shifts on every generation. Two or three photos from different angles are enough to stabilise it across a whole series of videos.
Recommended models
Seedream 5 Pro
Nano Banana
Flux 2 ProFrequently asked questions
Will my face stay recognizable?
Yes, that is what the Reference card does: your photos are sent to the model with every generation. The more varied the angles (front, three-quarter), the more stable the likeness.
Can the AI write the text inside the thumbnail?
Seedream 5 Pro handles short high-impact text very well. Keep it to 2-4 words: that is also what works best for click-through rate.
How much does a thumbnail cost?
A few cents per generation, with the price shown on the button before you click. The public calculator gives each model's exact rate without creating an account.
What size is a YouTube thumbnail?
1280 by 720 pixels, a 16:9 frame, which is the template's Image node setting. YouTube accepts files up to 2 MB, leaving plenty of room for a high-definition render.
Can I generate a thumbnail for free?
The workflow is free to explore, prompts included. Generating needs a free account, which opens with 80 credits, so sixteen thumbnails at five credits before paying anything.
Will my face stay the same across videos?
Yes, as long as you reuse the same Reference card: it lives in your library and is recalled with an @mention. That is what builds a channel identity recognisable at thumbnail size.
Is the text generated inside the image reliable?
Seedream 5 Pro handles short titles well, but always proofread. Another approach is to generate the background and face alone, then set the text in your own editor.
AI YouTube thumbnail maker
Use this workflowFree account, no card required. The workflow opens pre-filled in your canvas.