如何让 AI 视频风格统一:视觉设定集
如何让 AI 视频风格统一:做一页视觉设定集,包含风格段落、色调、每个地点一种光线逻辑和镜头焦段规则,然后在每条提示词中复用它。
在这一课中你将学到
- 如何为你的故事在写实和动画之间做出选择
- 如何写一段可以粘贴进每条提示词的风格段落
- 为什么 cinematic、8K 或 masterpiece 这类词什么也没描述
- 如何建立一套由四到六种颜色组成、各有含义的色调
- 如何为每个地点设定一种有动机的光线逻辑,以及各司其职的镜头焦段规则
- 风格帧测试,以及为什么道具设定图必须只展示道具本身
没有视觉设定集,每一次 AI 生成都会选出略有不同的视觉效果。这一课教你写一页决定,让影片中的每一条提示词都复用它,这样你的 AI 视频就能保持统一的风格。你要在写实和动画之间做出选择,写一段可以粘贴进任何提示词的风格段落,并把 cinematic(电影感)、masterpiece(杰作)这类空洞的词,换成具体、可检验的词。
然后你会建立一套由四到六种颜色组成、各有含义的色调,为每个地点设定一种有动机的光线逻辑,并用两三个焦段建立镜头语言。在 The Photo 中,你会看到四个镜头、三个焦段和两种运镜,全都出自这份设定集。这一课最后介绍风格帧测试、地点和道具的参考图,以及一个关于道具参考图、用惨痛代价换来的教训。
完整文字稿
英语旁白的文字稿。
0:04Before you generate a single shot, you have to decide what your film looks like. If you don't, every generation will decide for you, and every shot will pick a slightly different look. The fix is a visual Bible. One page of decisions about style, palette, light and lenses that every prompt in your film will reuse. In this lecture, we'll build the one for the photo.
0:29The first decision is the biggest, realistic or animated. It's not only a question of taste. Animation forgives a lot. Simplified faces and stylized worlds hold together more easily. Realism is unforgiving. Every face, every hand, every reflection has to convince. Lost Garden is an anime because the story is a fairy tale. The photo is realistic because it's an intimate family drama. Choose the look your story needs, and the one you can sustain for every shot.
1:03Then write your style as a paragraph you can paste into any prompt. Medium, era, texture, references. Here's ours. Photo -realistic night drama. 35mm film look. Shallow depth of field. Soft city bokeh through rain streaked glass. Warm sodium street lights against cool blue shadows. Subtle film grain. Notice it describes a look, not a story. The story comes from the script. The look comes from here.
1:37Compare that with cinematic, beautiful, 8K, masterpiece. Those words don't describe anything. They're the default every model already reaches for. Real style words are specific and checkable. A lens, a light source, a colour, a texture. If you can't check a word against a frame, cut it.
2:01Next, the palette. Four to six colours, no more. And give each one a meaning. In the photo, the beige coat and the burgundy scarf belong to Emily. The navy coat belongs to Thomas. Warm amber from the street lights stands for memory and warmth. A deep teal black fills the shadows. When the palette has a logic, the audience feels it without ever noticing it.
2:28Now, light. Give every location one motivated light logic and write it down. Inside Thomas' car, the only light comes from the street lights outside, through the wet glass. There is no light source inside the car. It's night and it stays night. Write that once, reuse it in every prompt and your shots will stop jumping from one lighting setup to another.
2:52Then, the lens language. Pick two or three focal lengths and give each a job. A wider lens, around 35mm, to show where we are. 50 for the medium shots, 85 for the close -ups where emotion lives. Add your rules for the camera. Eye level by default, a slow dolly in when emotion rises, a static frame when tension holds. These choices are what make a film feel directed.
3:19Here's how it plays out in the photo. Shot one, a medium two shot through the windscreen, 35mm, slow dolly in. Shot two, the same angle, static while he hands her the photo. Shot three, an insert on her hands, 50mm, static. Shot four, a close -up on her face, 85mm, dolly in. Four shots, three lenses, two movements. Every one of them is a decision from the Bible.
3:48Gather it all on a mood board. Six to ten images. Photographs, frames you generate yourself and stills from films you admire. Kept as references to study. Never to copy. In Screenweaver, this lives in the artistic direction section of the production canvas. An art style text node holds your style paragraph and an art style image node holds the look. Both feed every production reference in the project.
4:14Before you commit, run the style frame test. Generate three or four frames of key moments using only your Bible. The wide shot of the car, a medium on both characters, the close -up on Emily. Then check them against your page. Is the light motivated? Does the palette hold? If a frame drifts, fix the words in your Bible, not just the frame. A better sentence fixes every future shot at once.
4:39When you write prompts by hand, you paste the same Bible blocks every time. The style paragraph, the light logic of the location, the palette words. In Screenweaver, you don't paste them. The art direction feeds the production references. And in the storyboard panels to video pipeline, its text and image go straight to the scene video node.
5:02Your Bible also covers locations and props. Each location gets a reference image, like the interior of Thomas' car. Each key object gets its own sheet, like the photograph. And here's a lesson we learned the hard way. Our first prop reference showed the photograph on the back seat. In the generated film, Emily ended up sitting in the back. The model had copied the set along with the object. Since then, every prop sheet shows the object alone. Close up with no seats, no room, nothing that suggests a place.
5:38One more rule. Keep the Bible short. If your style paragraph runs longer than a few lines, the model starts picking and choosing. And so will you. A good visual Bible fits on one page. And you can read it out loud in under a minute. Everything on it is a decision. Everything that isn't on it, the model will decide for you.
6:01Your exercise. Write your visual Bible on one page. Realistic or animated. A style paragraph. A palette of four to six colours with their meanings. One light logic per location. And your lens rules. Then run the style frame test. The visual Bible template is in the resources. In the next lecture, we'll tackle the biggest challenge of AI filmmaking. Keeping your characters the same in every single shot.