这个爆火的 Gwen 手势特效，完整拆解给你看。

Transform the provided image into a polished semi-realistic 2D comic/anime illustration matching the provided comic-style reference.
Keep the original person recognizable and preserve the exact face, hairstyle, expression, body proportions, pose, clothing, accessories, and all visible details.
Preserve the original background, objects, composition, perspective, camera angle, framing, and lighting.
Only change the rendering style — do not redesign any content.
Style: refined semi-realistic 2D comic/anime illustration, clean expressive linework, detailed eyes, softly illustrated facial features, smooth illustrated skin, layered hair strands, natural anatomy, crisp contours, subtle cel shading mixed with soft gradients, rich fabric details, cinematic lighting, and polished digital illustration quality.
Transform the entire image, including the person, clothing, skin, hair, objects, and background, into one consistent illustrated style.
No photorealistic elements, no 3D CGI, no chibi proportions, no exaggerated anime anatomy, and no character redesign.
Original image = appearance, outfit, pose, and environment.
Comic reference = STYLE ONLY.

ChatGPT 端实际操作截图

Create a stylized 3D comic-book character using the reference image for the pose, body proportions, positioning, camera angle, and composition
Dress the character in an original futuristic white-and-pink superhero suit with sleek panel details, a white hood, and a flowing white cape.
Give the character a smooth white full-face comic mask with large expressive almond-shaped eye panels, bold black outlines, and vivid pink inner accents
Set the scene in a futuristic cyberpunk city filled with luminous pink neon, electric-blue accents, glowing signs, atmospheric haze, reflective surfaces, and dramatic rim lighting.
Keep the reference image's original pose and framing.
Style: vibrant stylized 3D comic animation, bold graphic shapes, saturated pink-and-blue colors, cinematic lighting, subtle halftone textures, dynamic contrast, and polished animated-film rendering.
Make the character and costume an original fictional design.

保留参考图片中角色的脸型轮廓。
将图片中的真实女孩转换成一个色彩鲜明的 3D 美式漫画风超级英雄角色，穿着白色与粉色相间的超级英雄战衣、白色兜帽、白色长披风，以及一个完全不透明的白色全脸面罩。
将角色整体完全风格化为 3D 漫画美学。不要保留任何真实的人脸特征，也不要尝试还原或重现真实人物的外貌。
保持角色原本的姿势、位置、身体比例以及整体画面构图不变。
将背景转换成一座赛博朋克城市，以粉色作为主色调，并加入蓝色作为点缀色。
整体采用鲜明的美式漫画风格视觉效果，使用大胆、明亮、高饱和度的色彩。

Use the provided reference image as the absolute visual source for the entire video. Keep the whole image visually identical to the reference image and animate ONLY the character.
The background must remain completely unchanged and static. Preserve the exact environment, objects, lighting, colors, perspective, composition, camera angle, framing, and every visible background detail from the reference image. Do not regenerate, redesign, replace, move, or animate any background element. No camera movement, zoom, pan, crop, or reframing.
Keep the character exactly identical to the reference image. Preserve the character's original visual style, face, hair, headwear or accessories, clothing, costume, footwear, body proportions, colors, materials, outlines, shading, textures, and every visible design detail. Whatever outfit or appearance exists in the reference image must remain unchanged throughout the entire video. Do not redesign, replace, add, remove, or modify any clothing or character detail while moving.
Use the provided reference video ONLY as motion data.
Transfer the hand-dance performance onto the character, matching the reference video's choreography, timing, and rhythm as closely as possible. Reproduce the hand gestures, finger positions, wrist rotations, arm movements, head turns and tilts, shoulder movement, torso sway, body bounce, weight shifts, and overall groove.
Match the left and right hands correctly. Preserve the original movement sequence, paths, speed, pauses, rhythm, and transitions. Keep the hands stable and anatomically consistent with five fingers.
Adapt the motion naturally to the character's original body proportions and clothing without changing the character design.
Do not transfer the reference video's person, identity, appearance, clothing, background, camera, lighting, or visual style.
The final result should look like the original reference image itself has come alive: the entire scene remains visually frozen and unchanged while only the original character performs the reference video's hand dance.
Reference image = ALL visual information.
Reference video = CHARACTER MOTION ONLY.