Teach you to use Dream AI to generate explosive short plays, zero-base mastery of the whole process of AI short play production

at presentDream AIVideo 3.0 came out of nowhere to give the AI VideoThe world has brought another small shock! One of the most prominent is its lens switching feature.

In the old days, each shot switch required us to re-create the image ourselves and then appear out of the video.

For example, in order to flesh out the characters, we'll do a 4-scene shot of the same character in the same scene, that is, we'll do 4 drawings by then, and then we'll use those four drawings to graphically generate the video, thus achieving the effect.

And with the Dream Video 3.0 model, you can just come up with the first picture, and then write the cue words for the 4 subplots in the cue words of the graph generated video, so that you can generate 4 subplot effects at once.

That's efficient, pulling straight through!

However, the current multi-camera switching for the consistency of the characters is still a problem, the following is the effect of the short drama I produced, the effect is still relatively rough, mainly to take you through the process of producing short drama, and learn to dream video 3.0 use.

Okay, here we go with a step-by-step teardown to take you through the build.

script for play, opera, movie etc

In terms of scripting, if we don't have any ideas yet and only have one direction, we can just talk to DeepSeek/ChatGPT more to get ideas.

DeepSeek I'm using Tencent Yuanbao here, the official website is here: https://yuanbao.tencent.com/

ChatGPT official website is here: https://chatgpt.com/

For example, we could ask this:

I would like to create a 3-minute AI skit on the theme of "Debut", please provide me with some ideas.

DeepSeek will give us a full script after thinking about it for a while:

We can see if it fits first, and just have it regenerate one if it doesn't.

However, you will also find that the scripts generated in this way are less controllable, sometimes too pie-in-the-sky, and larger compared to our expectations.

But that's exactly the effect we need, when we don't have an idea, looking at the script generated by DeepSeek, we slowly have a framework in our mind of what kind of short film we want to make and its storyline.

Once we've got that figured out, we'll be able to talk specifically to the AI and just let it help us refine it.

For example, we can write prompt words like this:

I want to make a short movie in the realistic genre with a plot line like this.

In the morning, a girl sits frozen by her bed, then it cuts to a city exterior scene where the outside of the city slowly changes from day to night, then it cuts back to the girl's scene where the girl displays a sad expression and then hides her face and cries. It then switches to a memory and a voiceover, "I'm Mei, I was a female model before," and the scene switches to Mei walking the carpet. NARRATOR: "I met him later, Xiao Shuai" and the scene switches to Xiao Shuai and Xiao Mei sitting opposite each other in a cafe drinking coffee. NARRATOR: "But, he's chosen other girls now", the scene blurs and only the backs of a guy and a girl are seen walking further and further away. NARRATOR: "No, I can't do this!" , the girl wipes away her tears, stands up, and runs out of the room. Narrator: "I can't be bound by the past", the camera switches to a city night scene, the girl runs through the streets Save the main content is different, help me touch it up!

Here's the full response, which still feels pretty well written:

Credits: Early morning shimmering light through the curtains. The girl sits quietly on the edge of the bed, in a trance, her eyes empty. The camera cuts out and a city street scene emerges. The street is bustling with traffic, moving from early morning to late afternoon, and then slowly sinking into night. Back inside: the room is dimly lit as night falls. The girl does not move, but finally looks down, tears slipping down her face. She covers her face, her shoulders shaking slightly. NARRATOR BEGINS, INTO A REMEMBRANCE SEGMENT: "I'm Mei, I used to be a model." The scene cuts to the red carpet, the lights flashing, Xiao Mei smiling confidently, her pace firm. "Then I met him - Xiao Shuai." The scene changes to a cozy café, the two of them sitting face to face, talking softly, unable to hide the smile overflowing from their eyes. "But he, in the end, chose another person." The camera blurred, only to see a man and a woman walking away side by side, their backs gradually disappearing into the depths of the lights. "I can't go down like this!" Mei looks up, wipes away her tears, takes a deep breath, gets up, and pushes the door open. CUT TO: A city night, neon lights flashing. Mei runs down the street, the wind blowing her hair, her steps becoming more and more determined. NARRATOR ENDS: "I shouldn't be stuck in the past. Now, I'm going to start over."

subscope

Once we've confirmed the script, we'll start creating the subplot by typing in the following cue words directly:

Help me break down the subplot and output markdown format, with the first column being the shot number, the second column being the scene, the third column being the scene, the fourth column being the runtime, the fifth column being the duration, the sixth column being the content of the scene, the seventh column being the narration, and the eighth column being the music.

Determining Characterization

Once we've confirmed the subplot, don't rush the figure yet, we need to keep the characters consistent, so we need to start with the character figure.

We'll start by having the AI come up with character descriptions, cue words:

Please provide me with a hint of Xiao Mei's character, I want her to be a Chinese woman, describe her in a paragraph, and provide both Chinese and English versions!

AI Response:

Xiaomei is a 25-year-old Chinese woman, a former model, with a tall, well-proportioned body and cool white complexion, and soft, delicate features. She wears her long, naturally flowing black hair, slightly curled at the ends and occasionally pulled back into a low ponytail, displaying a quiet and independent demeanor. She wears simple but fashionable clothes, often light-colored knitwear or soft long skirts, low-key but without losing a sense of high class. Her makeup is usually light, emphasizing her natural contours and look. Despite her elegant and detached appearance, her eyes hide layers of emotions: loss, struggle, longing and determination, like an unfinished story waiting to be discovered. She is unobtrusive but very expressive, the kind of female character who becomes more and more moving the closer the camera gets.

Xiaomei is a 25-year-old Chinese woman and a former fashion model. She has a tall, slender frame and fair, porcelain-like skin that highlights her refined and delicate facial features. Her long black hair falls naturally in soft waves, sometimes tied in a low ponytail, giving her a calm and composed elegance. Her style is minimalist yet sophisticated-often seen in light knit sweaters or flowing dresses, exuding quiet confidence. She wears light, natural makeup that enhances her subtle and sophisticated style. Her style is minimalist yet sophisticated-often seen in light knit sweaters or flowing dresses, exuding quiet confidence. She wears light , natural makeup that enhances her subtle expressions. While she appears graceful and reserved on the outside, her eyes reveal a complex inner world -sadness, longing, inner conflict, and eventually, determination. She's not one to demand attention, but the camera loves her-her presence deepens the longer you watch her, making her perfect for emotional, story-driven scenes.

Chinese prompt words, we can put namely dream to generate.

Instant Dreams official website: https://jimeng.jianying.com/

We choose "Image Generation", fill in the prompts, choose "Image 3.0" for the model, choose "SD 1K" for the clarity, and then click on "Generate Now":

Here we just choose the one we like:

The Little Handsome figures are acquired the same same way.

(i) Character consistency

Regarding the consistency of the characters, it's more dependent on the platform you choose to put out the images.

In Keyline, you can select "Keitu 1.5", then upload a reference image and select "Character Look":

In Dream, you can choose "Image Generation", then "Import Reference Image", and then "Portrait":

In versions 6.0 and 6.1 of Midjourny, you have the option of using "-cref" for portrait referencing, and in version 7.0, "-oref ".

However, since all the above tool references are single charts, the consistency is more general.

If you need better results, you can consider the SD training lora, or the "customized model" in Koring, which is a reference to multiple graphs.

Alternatively, we can use a bit of a face-swapping tool such as FaceFusion, the official website is here: https://docs.facefusion.io/

However, full consistency is not guaranteed by the above approach either, so we have another, more tricky approach.

It is that we are Asians, to Europe and the United States or Africa's human face is actually a bit of face blindness, at this point, even if you do not have to achieve 100% similarity, but also will be very similar, especially in the conduct of Vincennes video, this point is very important.

Output

(i) Mirror number 1

Since we've already created the shot list, we'll just let the AI make a picture based on each shot number, e.g., shot number 1, and we can write the cue word like this:

Please provide me with the cue words for Mirror 1 in both Chinese and English.

AI Response:

Of course you can. Here are the AI drawing prompts for Mirror 1 (Girl sitting on the edge of her bed in the morning, in a trance), in Chinese and English, suitable for generating realistic images.

🎯 Chinese cue words (suitable for national style writing / Chinese AI platform)

Realistic style, Chinese young woman, 25 years old, fair skin, long black hair falling naturally, sitting on the edge of the bed in a trance, wearing a beige knit sweater and baggy house pants, natural light coming in from the curtains in the early morning, the bedroom is decorated with simple warmth and the bedsheets are a little bit messy, the composition is quiet, the mood is low, cinematic, photographic style is realistic, high definition image quality

🎯 English Prompt (for Midjourney, GPT-4o, etc.)

Realistic style, 25-year-old Chinese woman with fair skin and natural long black hair, sitting blankly at the edge of a bed, wearing a cream knit sweater and loose lounge pants, soft morning sunlight coming through curtains, cozy minimalist bedroom, slightly messy bedsheets, emotional and quiet mood, cinematic atmosphere, high-resolution realistic photography, and a very nice and comfortable atmosphere. cinematic atmosphere, high-resolution realistic photography

I copy the Chinese prompt words since my platform for charting is i.e. Dream.

We choose "Picture Generation", then fill in the prompt words, choose "Picture 3.0" for the model, "SD 1K" for the definition, "16:9" for the picture ratio, then we click "Import Reference Picture": "Import Reference Picture": "16:9", then we click "Import Reference Picture": "16:9". "16:9", then we click "Import Reference Image":

We select "Portrait" and click "Save":

We then click on "Generate Now", you can get the effect:

(ii) Mirror number 2

We can start by looking at the cue words it provides in Mirror Number 2:

Realistic style, time-lapse images of urban exteriors, high-rise buildings, street traffic, crowd activities gradually reduced, the sky from the blue sky and white clouds slowly transitioned to the sunset orange and red, and then to the night neon lights on the state of natural and smooth changes in light and shadow, the shooting angle of overhead shooting or wide-angle telephoto, cinematic sense of composition, the atmosphere of the modern city, the sense of the passage of time is strong, high-definition image quality!

It this actually contains dynamic effects, which is not true, for which we have two ways:

1. First and last frames

We come up with a picture of the scene in the early morning, and then a picture of the scene at night for the first and last frames.

2、Tu Sang Video

All that is needed is a picture of the scene in the early morning, and then a cue word to control the picture and gradually turn it into a night effect.

Here, I'm using the second approach, so I just let it provide me with the early morning cue words:

Just come straight out with the early morning scene cue words and change to the night effect, which I generate using graphic born video

AI Response:

Realistic style, modern city in the morning, high-rise buildings, empty streets or a small number of commuters, the sky is white, the morning light softly spills down from the gaps in the high-rise buildings, the streets are slightly misty, the sky with a light blue and slightly orange tones, the glass curtain wall reflects the morning light, composition wide-angle or overhead shooting, the atmosphere is serene, the morning of the city awakening feeling, high-definition image quality, the sense of cinema!

Effect picture:

(iii) Mirror number 3

Since the characters and scenes of Mirror 3 are exactly the same as those of Mirror 1, we need to "intelligently reference" the generated image of Mirror 1 when we get the cue.

Tips:

Realistic style, Chinese young woman, 25 years old, fair skin, long natural hair, sitting on the edge of the bed in a dark bedroom, eyes sad, tears in the corners of her eyes or tears slipping down, wearing a housecoat or light-colored knit sweater, the room has only a weak bedside lamp or the night light of the city coming through the window, the background is bokeh, the picture is quiet and depressing, the emotions are introverted, cinematic composition, high-definition image quality

We select "Image Generation", fill in the prompts, and then click "Import Reference Image":

Select "Smart Reference" and click "Save":

We will click "Generate Now" to get the following effect:

All other diagrams can be generated as above.

publish a video

(i) Shot 1

Image content: girl sitting on the edge of the bed, hands hanging down in her lap, eyes staring blankly

Since it's a character freeze effect, we can just use the image without having to come up with a video.

(ii) Shot 2

Content of the image: the city changes from day to night, traffic, crowds, sky color change

A delayed change effect is used, so we upload the early morning image and then just control the change inside the cue word.

Tips:

The smoke fades and the sky brightens to midday, then the sky darkens again to evening

(iii) Lens 3 + Lens 4

Content: camera 3: the girl sits in place, her eyes sad, tears slipping from the corners of her eyes, camera 4: the girl hides her face and sobs, her shoulders gently trembling

We've merged the camera switching features of Dream Video 3.0, for example, directly into one.

Tips:

Camera 1: The girl's eyes are sad, tears slip from the corners of her eyes, Camera 2: The girl hides her face, her shoulders gently trembling

(iv) Lens 5

What: Shot 5: Mei walks confidently down the red carpet in a gown

Here, we can give the character multiple camera switches so that it appears to be more specific.

Tips:

Female model walks forward 3s camera cuts to female model's face, female model walks forward 6s camera cuts to female model's right side face close-up, female model walks forward

(v) Shot 6

Content: Shot 6: Xiao Mei and Xiao Shuai sitting by the window, smiling at each other, warm atmosphere

For opposite effects like this one, we can cut down the front and back shots.

Tips:

Man and woman chatting at each other's eyes, 2s and 4s, each cut down to front and back shots

The effect of the lens used in the back is the same as in the front, so I won't repeat it here.

VII. Editing

In terms of editing, it's mainly just a matter of importing the images and videos that came out before and then sorting them. The detailed process is then explained in detail in the live broadcast.

When it comes to soundtracks, we first need to understand the story well.

My feeling here should be from sadness to calmness to passion. So I open Suno and fill in the cues directly:

Suno official website: https://suno.com/

From sadness to calm, to passion

Thank you all for seeing this!

I'm bursting at the seams with 5,000 words, and I'm sharing them gratuitously in the hope that they'll be helpful to you!