A fully AI-generated character — moustache, brown suit, blond bob — gets dropped into real viral clips in place of whoever was originally there. It hit 100 million views in its first week. How it actually works, which tools make it, and why it's not the same thing as a deepfake.
A consistent, fully AI-generated character — moustache, brown suit, blond bob, usually called "Jean Phil" — gets dropped into an already-viral video clip in place of whoever originally appeared in it, with the clip's real motion, camera movement, and color grading left intact around him. The effect reads as a surreal, deadpan cameo: familiar internet-famous footage, but with an invented character calmly going about the original action. It emerged in mid-September 2026, reportedly hit 100 million views within its first week, and has since spawned at least one rival character and a wave of copycat formats using the same swap-a-character-into-real-footage technique.
Worth flagging before you try this: the source clips being edited are almost always someone else's existing viral footage, not original material. Reusing another creator's video as the base layer for a character swap sits in a legal and platform-policy gray area that varies by platform and by how transformative the result is judged to be — this isn't a settled question, and we're not offering legal advice here. If you want to build something in this style without that risk, the safer version is applying the same technique to your own original footage rather than someone else's viral clip.
The repeatable structure behind a character-swap clip:
The joke's entire mechanism depends on the viewer recognizing the original clip — the humor is in the contrast between familiar, real footage and the calmly out-of-place invented character, not in the technical novelty of the swap itself. A technically perfect swap into an obscure clip nobody recognizes falls flat for the same reason a reference joke falls flat on someone who doesn't know the reference.
Creators most often cite Seedance 2.5 and Kling for the core swap-and-regenerate step, since both handle motion-preserving edits of existing footage well. Hedra is specifically credited for keeping the character's face and expression consistent across multiple source clips — the thing that breaks a recurring-character bit fastest is the character subtly looking like a different person in each new video. None of this is a single-tool, one-click workflow; expect to combine a character-consistency tool with a video-editing one.
Standard per-clip pricing for whichever tool you use — no special pricing tier for this trend specifically, though expect to pay for generation on both the swap tool and any separate consistency tool if your workflow uses more than one. See our full AI video generator pricing comparison, or the cheapest way to try it if you just want to test the format first.
Once you've got a clip you like
The generated clip comes with its full background baked in — no built-in way to export just the character on its own. If you want to turn a frame into a sticker, thumbnail, or composite it into something else, that's a background-removal step done after generation. RemoveGifBG processes AI-generated MP4s frame by frame and hands back a transparent result — see the AI video background remover guide.
Related: How AI videos are made · Real deepfake controversy, for comparison · The AI baby dance trend · The AI ASMR trend
Jean Phil is a recurring, fully AI-generated character — a mustachioed figure in a brown suit with a distinctive blond bob — that creators drop into existing viral video clips in place of whoever originally appeared in them, keeping the same motion, camera work, and color grading intact. It emerged in mid-September 2026 and hit 100 million views within a week, spawning rival characters and copycat formats. Unlike a deepfake of a real person, Jean Phil is an entirely invented character — no one's actual likeness is being used or impersonated.
No, and the distinction matters. A deepfake replaces footage with a real, identifiable person's actual likeness, usually without their consent — that's the category behind genuinely serious controversies elsewhere in AI video (see our coverage of the Grok Imagine situation, for instance). The Jean Phil trend instead replaces the original footage's subject with a wholly invented, fictional character who doesn't correspond to any real person. The technique — swap a consistent character into existing video while preserving motion and camera work — is similar to deepfake tooling under the hood, but what's actually being generated is a cartoon-like original character, not a fabricated likeness of someone real.
Creators most often cite Seedance 2.5 and Kling for the core character-swap generation, with Hedra credited specifically for keeping the character's face and expressions consistent across different source clips. No single tool handles the whole pipeline end to end — this is a multi-step workflow (source clip selection, character reference, generation, sometimes manual compositing), not a one-click effect.