I made it into a horror game AL COMICPerformance。
Its game sense comes from first person lenses, from perspective switching, and later superimposed interactive tips and dialogue boxes. It's really hard to get these things set up in the same story。
THIS PRODUCTION ALSO TOOK THOUSANDS OF POINTS AND TOOK A NUMBER OF TURNS. HERE'S A COMPLETE BREAKUP OF SCRIPTS, PERSON AND SCENE ASSETS, VIDEO GENERATION, CAMERA REPAIR, AND GAME UI AND SUBTITLES。
I. Looking for an anomaly from a familiar day
To find an entry point in your own life
The first step is to write the script。
We don't need to build a complex world view for a minute or two of short filmsA complete story can be formed by clearly identifying the person, the environment and the underlying assumptions, moving on around an anomaly and finalizing a reverse.
If there are no ideas for the time being, they can start with the occupation or living environment of which they are familiar。
Think about it:What happens when something happens suddenly in a familiar day
For example, a teacher's class suddenly came to a strange transfer student
A lifeguard was locked in a swimming pool and strange creatures were found in the pool
A driver drives in the morning and goes back to the same road, no matter what
A housewife found that the present husband, while he looked the same, was becoming more and more strange..
These “on a day-to-day basis anomalies” can be the starting point of the story。
Once we have the initial idea, we can use itDeepSeek, GPT or bean bagTo expand the creativity into a script。
At this point, we'll be wondering, which model should we use
The tools I used this time, in general terms, are:
Bean bag job:there's a lot of skills in there, and you can choose the skills of short play, and you can combine them with better barriers on the market, and you can do water-lined shorts. in this experience, the drama needs to continue。
GPT:MORE ALMIGHTY, MORE CREATIVE AND BETTER SCRIPTED LINES. THE LINES ARE MORE FLUID AND LIVELY THAN SOYBAGS AND DSS。
This time, the usual division of labour was followed: during the creative stage of the script, I used to use Astra's high-level reasoning file, medium- or high-intensity sal when writing scripts and changing lines。
Deepseek:
i usually call deepseek's model in workbook
taken together, the model of ds is also very good in the script, similar to that of gpt. 5.6, and often gives me a more surprising reverse。
However, it is not unusual to write character lines。
you can also choose to add a number and call some of the short scripts in the workbooky so white can have more options。
WHATEVER MODEL IT IS, TELL IT THE STORY, THE TARGET TIME, THE DESIRED ATMOSPHERE AND THE REVERSE DIRECTION, THEN GRADUALLY ADJUST THE CHARACTER'S BEHAVIOR, THE PLOT RHYTHM AND THE ENDING TO MAKE THE SCRIPT MORE SUITABLE FOR SUBSEQUENT AI VIDEO PRODUCTION。
WE'LL NEED MANUAL FINE-TUNING AND FINE-TUNING WHEN AI FINISHES THE PLAY。
TWO: FIRST, YOU SET THE PERSON, SCENE AND UI ASSETS
character (in a play, novel etc)
When the script is finished, don't rush to create pictures. We need to determine the visual style of the entire short film。
Both the person and the scene must be organized in the same direction so that the resulting images can remain unified。
This time I wanted to feel like I was getting into a horror game, so I putVisual directionSet THREE-D, THREE-A GAME QUALITY。
Once the direction is set, a third view of the person and a map of the scene can be generated from the script。
And here's some of the pictures that I've produced, and you can look at them as a whole。
It's the end of the figureDreamseedream5.0Flash
The prompt words are as follows:
ASIAN FEMALE STUDENTS, 16 YEARS OLD, PONYTAILS, STRONG FILM PARTICLE SENSE, SUPERREAL, FANTASY ENGINE 5, CG STYLE, 3D COUNTRY-WIDE, HALF BODY
andoh, ohdream 5.0 placeThat's how it feels。
IF YOU'RE LOOKING FOR A MORE 3D STYLE, YOU CAN ADD IT TO THE HINTTHE BJD.
ASSET DESIGN IS THE TOP PRIORITY FOR AI VIDEO PRODUCTION。
Before officially generating videos, you can try more models to compare the effects and carefully sharpen the assets of people, scenes and props。
Especially for peopleAt this stage, it would be preferable to define the appearance, hair, clothing and overall style. Because the next spectroscopy and video will be built around these settings, half-modified, often involving multiple lenses and even regenerated。
A little more time was spent in the prior period to complete the assets, so that the later production would be smoother and would reduce back-to-work。
Once the overall image of the role has been determined, the details of the front, side, back, clothing, hair and body ratio have been completed。
I doGPT's imageThe graphic function of the character three view。
If you do not have a GPT account, you can also use any platform, such as libtv。
The three-view hints are as follows:(16:9/21:9)
Generate a three-dimensional view of the body and a facial close-up. The top left is 1/3 full of super-large facial features, 2/3 right, positive view, side view, back view, two or eight body parts ratio, pure background
When the third view is ready, someone needs to change。
If you need to change the man's costume, we can do it with a banana model
The image of the person is more visible when replaced with an Image model。
NPC DESIGN. I'VE CONSULTEDThe Next DayThe atmosphere of these horror games makes the role faceless。
In an ordinary school environment, there are people who do not have five officers but still act as usual, and this contrast itself creates a sense of anxiety。
The role ' s clothing and postures can be kept on a daily basis, with abnormal focus on the face。
At the same time, the basic scenes to be used in the script are prepared. First determine what the space looks like, then create different lenses around it。
take
I'm using the same thing for scene assetsimage modelAll we have to do is describe the image and the style。
For example, the subject of school terror can be written like this
3D 3A GAME DREAM CORE CAMPUS TERROR GAME QUALITY, 16:9 SCREENS, DAYTIME, LOW SATURATED AND COLD GRAY, BLACK AND SOFT LENS EFFECTS, REAL MATERIALS, SOFT NATURAL LIGHT。
When the basic scene came out, I used LibTV to change the angles, and I continued to generate other views in the same space。
This step is well suited to the picture of follow-up needs, such as looking from the corridor to the classroom, from the door to the room, or looking back at the position just passed。
When I create different anglesOr in a hintKeeps a description of the same style in which the material, light and color are as uniform as possible。
For example, the subject of school terror can be described as follows:
THE 3A GAME OF DREAM NUCLEI SCHOOL TERROR GAME QUALITY, DAYTIME, LOW SATURATED, COLD AND GRAY, DARK LENS EFFECTS, REAL MATERIAL, SOFT NATURAL LIGHT。
UI Design
When the scene is created, we can also select some of the images and add them by redrawing themUI ELEMENTFor example, interactive tips such as “into class” “on the phone” make the picture more game-oriented。
YeahGENERATE A SINGLE SET OF UI BUTTONS, ICONS AND INTERACTIVE LABELS, post-cut, bottom-up, then fold into the video。
This allows for flexibility in adjusting their location, size and timing to the needs of the lens。
III. Choosing video models around the current style
Once the material is ready, you can get into the video generation。
This step suggests that you start with the same set of materials to test different models and see which one is better suited to the style you want。
THE MORE EXPENSIVE THE MODEL IS, THE BETTER IT WORKS, THE MORE IMPORTANT IT SUITS YOUR IMAGE AND NARRATIVE NEEDS。
This time I tested H3, SD2.5, SD2.0, Wan3.0 and other models.
00:15
What I want around this short film 3D, 3A GAME QUALITY AND ONE-PERSON PERSPECTIVEMy actual feeling is:
• SD2.5:IT'S GOOD, BUT IT'S MORE VISUAL, AND THERE'S STILL A DIFFERENCE BETWEEN THE 3A QUALITY AND CG FEELING I WANT。
• SD2.0:Combined performance is still good, but the adherence to the hints is common, as is the consumption of points。
• H3:The overall effect is yet to be achieved, but compliance with the hints is also not ideal, especially with first-person lenses, which still lack a sense of proxy in some games。
• Wan3.0:THE 3D TEXTURE OF THE REFERENCE MAP IS BETTER, AND THE GRADE CONSUMPTION IS RELATIVELY LOW, AND IS A MODEL THAT MEETS MY NEEDS。
This is mainly about the material and style of this short film. When you make it, you can also test a few representative shots and then determine the main model to follow。
Taken together, I used mainly the Wan3.0 model, supplemented in part by SD2.5。
The basic process of video generation is similar to the normal AI video production, choosing the Wan3.0 model, choosing the ratio, length, resolution, etc。
但The first person said that the perspective required a more detailed description of the camera, especially by stating: “Where am I, what am I looking at, how do I move next?”。
We can choose the frame of reference first, and then make it clear in the hint: where the lens begins, where it goes, who's in the picture, and what happens next。
The perspective can be written POV, THE FIRST PERSON'S SUBJECTIVE VIEWI DON'T KNOW. BUT IT IS NOT ENOUGH TO WRITE A “POV” AND TO DESCRIBE THE ACTUAL BEHAVIOUR OF THE LENS. FOR EXAMPLE:
keep subtitles free, avoid creating any text or subtitles, do not generate logo, do not produce watermarks。
The first perspective pov of the whole journey generates only one shot of a single long shot. The first person, who woke up in a chair, was sitting in front of her, sitting in front of her, and she looked around and slid back to her, looking back at her, looking at her, looking at her hand (the lady's sleeve in a blue school uniform, blue pants, white shoes), and suddenly the bell rings from a distance. There's a teacher in the school who's coming in the camera, standing in front of the camera, ready to start talking to the master. Interact ui click teacher。
pov view of game quality Fantasy Engine 5, 3A game CG style. No music, no sound
Clear vision, direction and sequence of actions will bring the lens closer to the images that the character sees and to create a sense of proxy in the game。
In this way, perspectives, personality relations and movement orientation are all based。
IN ADDITION TO THE FIRST PERSON'S POV SHOT, WE CAN ALSO INSERT SOME THIRD PERSON'S OBJECTIVE SHOT。
The third person referred to the perspective as usually referring to the play-time。
Through lenses such as panorama, vertebrae, the audience is able to see the person's environment and the person's relationship to space。
When the picture is transposed from the first to the third, it also temporarily emerges from the subjective view of the role and presents the scene and character in a more objective sense。
The point of view is removed from the eye of the role and moved to the high view, creating a viewing effect similar to that of God and making the change of perspective clearer。
However, if the first perspective is changed to the third person's claim, a single “third person's perspective” is sometimes unstable。
I'm going to further define the composition of the character in the picture, and I'm going to make it clear that the camera is outside the lead。
For example:
Push the camera slowly, take the camera over, the lead person and the teacher appear in the picture at the same time: the main person stands for the future, back-to-back; the teacher is on the other side of the corridor, looking at the main person. The image is a complete picture of the distance between two people。
iv. Patching the lens with a front-end frame
Video production makes it difficult to generate first-view material, which often requires a large number of cards, but which often do not produce good results。
In this case, I'm going to fill in a little transition video with a filamental frame。
This material, for example, feels good, but suddenly it is drawn。
00:00 / 00:15
This will be done by:
• Interception requires modification of the frame of the video。
• Intercept the frame of the appropriate video。
• Combination of lenses with a filamental lens。
• The resulting transition clips are shown with the deletion of duplicate head and tail frames and the checking of the linked clips。
If the space between the two images is too different, don't be reluctant to connect. It is often more natural to adjust the material first or to add an intermediate perspective。
V. ADD UI AND GAME SUBTITLES LATER
After the base was finished, it ended up in the editing environment。
Regular editing, with subtitles and hint interfaces。
THE HINT INTERFACE ALLOWS YOU TO PUT READY UI IMAGES, SIMPLY FOLDED AND CUT, INTO THE VIDEO。
Depending on the scenario, a short reminder such as “Enter the classroom” “See the Glory List” can also be added。
The location and time when the hint appears must correspond to the camera action. When the camera comes close to the door, “to the classroom”, the audience will understand its relationship to the current image。
For subtitle designI'm using a more direct method:
• Copy the original video to the upper orbit and align it with the lower material。
• Use a masked frame to map out the areas to be subtitled, repositioned and added plume and shadow effects。
• Add a vague effect to this upper layer of material, which only obscures the background of the subtitle area。
• Adding text above the fuzzy area, adjusting fonts, and changing fonts to pixels。
In this way, the colours and changes of the original image are preserved behind the subtitles, while at the same time being able to draw a layer from the background and create a similar effect as the game dialogue box。
The fonts, margins and subtitles are also as uniform as possible, so that the entire film looks like the same game interface。
Start with a little story
AND HERE, AN AI SHORT PLAY WITH A GAME INTERACTIVE TIP IS ALMOST DONE。
THE GROWING NUMBER OF PEOPLE WHO ARE NOW DOING THE AI VIDEO, THE AI SHORT PLAY, WHICH COMBINES SHORT PLAY NARRATIVES AND GAME EXPERIENCES, REMAINS A PROMISING DIRECTION。
this kind (of)A lightweight story,It's not that goodIt's just an unusual setup, through perspective, sound and interactive tipsInteresting interactionLet the audience step into the story。
When you run the process, you can try more on the subject and find your own way of expressing itIt's a form of film integration worth trying.
FROM SCRIPTS, PEOPLE AND SCENE ASSETS, TO VIDEO GENERATION, TO CAMERA REPAIR, TO UI AND LATER CLIPPINGS, THE WAY I TRY TO FIND OUT。
YOU CAN START WITH A LITTLE STORY FOR A MINUTE OR TWO AND FINISH THE FIRST ONE. THE PROCESS IS STUCK, AND IT IS RE-ADAPTED: THE STORY DOES NOT CREATE A SCRIPT, THE PERSON CHANGES HIS ASSETS INCONSISTENTLY, THE CAMERA MAKES UP THE TRANSITION, AND THE GAME FEELS INSUFFICIENT TO EXAMINE THE PERSPECTIVE AND UI。