用 AI 重建《请以你的名字呼唤我》的壁炉场景

· Flick 内部影片制作人 · LinkedIn

Aug 2026

The fireplace scene in Call Me by Your Name was one of the scenes that I just couldn’t forget after seeing it. I wanted to recreate the blocking, framing, and camera movement that Guadagnino brilliantly implements here. So here’s how I did it, for free, with AI.

If you haven't seen it in a while: the film's final shot is one unbroken take. Elio sits in front of the fire, the family sets the table behind him, the credits roll over his face, and nothing else moves. That's exactly what makes it the perfect first scene to rebuild: one locked camera, one subject, one light source doing all the acting.

  1. Find the ideal stills from classic film scenes.
  2. Edit the image with Nano Banana Pro to create a realistic human.
  3. Generate the video.
  4. Detach the video and audio, then rebuild the sound.

步骤 1:选择你的画面

找到正确的静态图像。找一个镜头不移动、主体在构图中单独出现、光线有一个清晰来源的画面。

有一点需要明确:这个画面是一个参考,而不是素材。它永远不会出现在你的输出中。它的存在是为了让你能够回应它,就像摄影指导回应光线图一样。

Reference still from the fireplace scene pinned on the Flick canvas as the composition and lighting guide
Reference still from the fireplace scene pinned on the Flick canvas as the composition and lighting guide

Rebuild your favorite scene

One still, one new face, one short clip.

步骤 2：用新人物重建静帧

现在开始重建。在 Flick 中，这是针对你的参考进行的 Nano Banana Pro 编辑工作。描述你的人物，然后像你导演现场一样一次修正一件事：我把沙发改成了蓝色，并替换了我的主角。

The rebuilt still: a new person by the fire on a blue sofa, same composition and firelight as the reference
The rebuilt still: a new person by the fire on a blue sofa, same composition and firelight as the reference

Step 3: Generate the video

使用图像到视频功能，将你完成的静帧固定为第一帧，并将运动提示词写成一系列"不要做什么"的列表。镜头锁定。不缩放，不漂移。火焰是唯一的真实运动。人物呼吸；他们不表演。五秒钟，一次完成。

这种克制就是整个技巧的关键。原始场景之所以著名，是因为除了火光在脸上的变化，什么都不动。模型产生的每一个额外动作（转头、镜头推进、室内光线闪烁）都会打破这种魔力，所以要像花钱一样控制运动，几乎不花。

Image-to-video setup in Flick with the rebuilt still pinned as the first frame and a locked-camera motion prompt
Image-to-video setup in Flick with the rebuilt still pinned as the first frame and a locked-camera motion prompt

Step 4: Rebuild the sound

Generated clips rarely land audio timing, and this scene lives on its sound: fire crackle, room tone, the family somewhere behind. Detach the generated audio, then rebuild it in layers and sync the crackle to the visible flicker by hand, even roughly. Score it with your own music; the original's needle drop belongs to the original. Five seconds of picture with honest sound reads as cinema. The same trick is step four of the method Zoey published with her original fireplace recreation back in February.


成果

Look at the frame again. Long dark hair, red sweater, blue sofa, the hearth burning at the edge of a cool, book-lined room. Not one pixel comes from the film, and you knew what it was before you read a caption. That makes for a great homage.

When you want a second scene, the same four steps produced remixes of In the Mood for Love and the Atonement running shot. Pick a scene with a locked camera first; save the tracking shots for your third.

常见问题

AI 能重建真实电影中的场景吗？

作为重建的致敬，可以：相同的构图、光线和镜头语法，画面中是一个新人物。这就是本文中发布的 Flick TV 重建背后的四步法。

我需要什么工具？

这个片段使用了 Nano Banana Pro 进行静帧编辑，以及 Flick 的图像到视频模型来实现运动，全部在一个画布上完成。静帧承载场景；视频处理只让火焰动起来。

为什么壁炉场景如此适合这个方法？

锁定镜头、单一主体、一个有动机的光源。这三个特性正是图像模型能可靠再现的，这使得这个场景成为在你尝试任何带镜头运动的内容之前的最佳首个项目。

为什么不直接在提示词中写导演的名字？

改为描述美学特征：温暖的单一来源壁炉光、凉爽的意大利室内、胶片颗粒感、静止的固定画面。模型执行描述的光线效果比使用名字要好，而且对在世导演的风格点名请求会让内容过滤器变得敏感。