FatThumb's editor applies a plain-language instruction to an existing thumbnail across five scoped modes. The original is reproduced exactly except for the change you asked for, and every edit lands as a new version you can compare and revert.
How it works
Choose the scope of your change: Modify for general edits, Text for wording and typography, Style for the visual aesthetic, Emotion for the facial expression, or Face for face-region changes with an identity lock.
Type what you want — "make the hoodie red", "change the headline to FINAL ROUND", "make me look shocked". The editor anchors on your existing thumbnail and applies only the delta you described.
Each edit produces a new version while the original stays in your history. Compare versions side by side, star the keeper, and download it as an exact 1280x720 PNG ready for YouTube Studio.
Features
Modify, Text, Style, Emotion, and Face. Picking a mode tells the AI exactly what it is allowed to touch, so a wording fix cannot drift into a face change and a restyle cannot move your layout.
Every edit instruction opens by anchoring on your existing thumbnail: reproduce it exactly except for the described change. Composition, subject, colors, and text outside your request stay put.
Face mode uses your Person profile photos as the authoritative identity reference, and the strict-likeness toggle keeps the face exactly as-is — no beautifying, aging, or substituting while the rest of the edit happens.
Edits never overwrite the original. Each one is saved to your version history, so you can branch from an earlier direction, revert a misfire, or download any past version at any time.
View up to four variations next to each other in the compare view. Judging candidates in one screen beats flipping between tabs and trusting your memory of the last one.
If an edit fails or is refused by content moderation, the generation stops and your credits are refunded. There is no silent retry and no charge for output you never received.
FAQ
Five scoped modes cover the common cases. Modify handles general changes like swapping a background object or adding a prop. Text changes only the wording and typography. Style changes the visual aesthetic — colors, lighting, rendering — while keeping the subject and layout. Emotion changes only the facial expression. Face applies face-region changes with an identity lock from your Person profile.
No. Every edit instruction is anchored on the existing thumbnail with an explicit directive to reproduce it exactly except for the described change, and the mode you pick constrains the scope further. A Text-mode edit, for example, carries an instruction to preserve the subject, composition, and colors exactly.
In Face mode, your Person profile photos are attached as the authoritative identity reference, and the AI is instructed to treat them as the source of truth for your face. With the strict-likeness toggle on, the instruction is to keep the face exactly as-is — same features and proportions, with no beautifying, aging, or restyling.
Nothing is lost. The edit is saved as a new version alongside the original, so you can compare them, go back to any earlier version, or run a different instruction from whichever version you prefer.
A failed or moderation-refused edit does not produce an image and your credits are refunded automatically. FatThumb does not auto-retry refused edits — you adjust the instruction and decide whether to run it again.
Iteration generates fresh variations from a prompt — up to four new directions per pass that you compare and star. The editor works on a thumbnail you already have: it makes one targeted change and keeps everything else fixed. Most creators use both — iterate to find the direction, then edit to perfect it. See the iteration and version history page for the first half of that workflow.
Yes. FatThumb includes a pixelate/anonymize option for legitimate privacy needs — for example, a guest or bystander who should not be identifiable. It is built for anonymity, not for misrepresenting who appears in a video.
Generate your first 5 thumbnails free — no card, no designer, your face consistent from the first run.