An end-to-end recyclable game effect automatic generation and processing method and system

Abstract

This invention relates to an end-to-end automated generation and processing method and system for loopable game special effects. It constructs a regularized game special effects video dataset, adjusts the parameters of a pre-trained video generation base model to obtain a vertical game special effects generation model, acquires user input information and inputs it into the vertical game special effects generation model to generate game special effects videos with low-brightness backgrounds, processes the generated game special effects videos, and outputs a sequence of frame images adapted for game engine rendering. The system includes a user end and a server end. The server end includes a prompt word processing module, a model routing and generation module for storing the model and executing the method, and a post-processing module. This invention significantly reduces the production cost and usage threshold of professional game special effects, ensures high availability and standardization of the generated results, smoothly connects the beginning and end of the generated sequence of frames, and ensures that the generated special effects strictly conform to the logical constraints of game design while taking into account creativity and controllability, thus constructing an efficient data production pipeline.

Description

Technical Field

[0001] This invention relates to the technical fields of card games, board games, or roulette games; indoor games using small moving objects; video games; and games not included in other categories. In particular, it relates to an end-to-end automated generation and processing method and system for loopable game effects in the fields of artificial intelligence and computer graphics. Background Technology

[0002] With the rapid development of generative artificial intelligence (AIGC) technology, its application in image and video generation is becoming increasingly widespread. In the game production industry, visual effects (VFX) are a key element in building game visual feedback and immersion. However, traditional game VFX production relies heavily on particle systems, hand-drawn sequence frames, or 3D software rendering, resulting in complex processes, long cycles, and high labor costs.

[0003] While current widely used text-to-video (TTV) models, such as Sora and Runway, possess powerful generation capabilities, they primarily target streaming content like movies and short videos, focusing on generating general-purpose videos with complex backgrounds and coherent narratives. However, the game industry has extremely specific professional requirements for special effects materials. These include the need for clean backgrounds (usually transparent or pure black) for game engine compositing, smooth frame transitions for infinite looping, adherence to specific frame rate and resolution specifications, and precise control over the motion trajectory of effects to match game logic. Directly applying general video generation models to game special effects production faces significant challenges: (1) Existing general models often generate videos with complex backgrounds, making it difficult to generate glowing effects on pure black backgrounds, which leads to difficulties in later image cutout, with residual noise at the edges, and the resolution and aspect ratio are not fixed, making it impossible to directly obtain the pure background (or Alpha channel) sequence frames required by the game engine. (2) The first and last frames of the linear video generated by the existing general model cannot be connected naturally. This cannot meet the requirements when the first and last frames of the video are perfectly connected to achieve infinite loop playback for special effects such as idle, halo, and continuous skills in the game. If the video is replayed after it ends, there will be obvious jumps (flickering), which will ruin the game experience. (3) It is difficult to accurately control the physical trajectory of special effects based solely on text prompts, such as the parabola of bullets and irregular magic trajectories; (4) The entire process still relies on users to have superb prompt word engineering skills and perform tedious post-processing, including but not limited to PS retouching, frame extraction, alignment and other secondary processing. It has failed to achieve end-to-end automated production, and the production threshold is still very high.

[0004] Therefore, existing technologies suffer from technical problems such as low efficiency in game special effects production, incompatibility between general AI-generated content and game industry standards, difficulty in achieving controllable loops and precise motion trajectories, and insufficient automation. There is an urgent need for an automated solution optimized for the vertical field of games. Summary of the Invention

[0005] This invention solves the problems existing in the prior art and provides an end-to-end automated generation and processing method and system for loopable game special effects. It utilizes video generation models and large language models (LLM) to work together to realize the automated production of game special effects materials.

[0006] The technical solution adopted in this invention is an end-to-end automated generation and processing method for loopable game special effects, which constructs a regularized game special effects video dataset, and adjusts the parameters of a pre-trained video generation base model based on the dataset to obtain a vertical game special effects generation model. Obtain user input information, input the vertical game special effects generation model, and generate a game special effects video with a low-brightness background. The low-brightness background is a monochrome background and the brightness value is lower than a preset threshold. The generated game effects video is processed to output a sequence of frame images adapted for game engine rendering.

[0007] Preferably, the original special effects video footage is obtained, the background is uniformly processed to black, the video footage is spatiotemporally normalized, the processed video footage is classified, and data augmentation is performed within the same category to obtain a regularized game special effects video dataset.

[0008] Preferably, descriptive text is generated from the regularized game effects video dataset using semi-supervised automatic annotation.

[0009] Preferably, the parameters of the pre-trained video generation pedestal model are adjusted using a low-rank adaptive technique. This is achieved by injecting a trainable low-rank matrix into the attention layer of the pre-trained video generation pedestal model and freezing the original parameters of the pedestal model for training, while only updating the low-rank matrix parameters.

[0010] Preferably, when generating the game effects video, the user's input is expanded into professional prompt words through a large language model; based on the semantics of the expanded prompt words, a matching model is automatically selected from multiple vertical game effects generation models of different styles for generation.

[0011] Preferably, when generating the game effects video, the features of the first frame of the generated sequence are injected as constraints into the generation process of the last frame, or the end of the video sequence is filled in using video restoration technology with the first frame as a reference. The generated video has smooth frame transitions between its first and last frames, and plays in a loop.

[0012] Preferably, when the user inputs information including motion trajectory information, the motion trajectory information is converted into a spatial control signal when generating the game special effects video, guiding the generated special effects subject to move along the specified trajectory.

[0013] Preferably, the processing of the generated game effects video includes color filtering rendering optimization and automated frame extraction. Color filtering and rendering optimization processes generate videos with pure black backgrounds. Automated frame extraction processes sample game effects videos into multiple static images at set time intervals and arrange and stitch them together into a sequence of frame images.

[0014] Preferably, in the game engine, a color blending mode is applied for rendering, making the pure black background completely transparent, and the bright parts of the special effects are superimposed with the brightness of the game background.

[0015] An end-to-end automated generation and processing system for loopable game special effects, comprising a user terminal and a server terminal connected via a communication interface; The client is used to receive information input by the user and send it to the server, and to receive and display the results generated by the server. The server-side includes: The prompt word processing module is used to call the large language model to enhance the received information; The model routing and generation module stores a pre-trained vertical game effects generation model, which is used to generate game effects videos with low-brightness backgrounds based on the enhanced information and the vertical game effects generation model. The post-processing module is used to process the generated game effects video and output a sequence of frame images adapted for game engine rendering. The model routing and generation module and the post-processing module execute the end-to-end loopable game effects automated generation and processing method.

[0016] This invention relates to an end-to-end automated generation and processing method and system for loopable game effects. It involves constructing a regularized game effects video dataset, adjusting the parameters of a pre-trained video generation base model based on the dataset to obtain a vertical game effects generation model, acquiring user input information, inputting it into the vertical game effects generation model, and generating game effects videos with a low-brightness background. The low-brightness background is a monochrome background with a brightness value below a preset threshold. The generated game effects video is processed to output a sequence of frame images adapted for game engine rendering. The system includes a user terminal and a server terminal. The server terminal includes a prompt word processing module, a model routing and generation module for storing the model and executing the method, and a post-processing module.

[0017] The beneficial effects of this invention are as follows: (1) Significantly reduce the production cost and usage threshold of professional game special effects. Users do not need to have a deep art background or complex prompt word engineering capabilities. They can produce professional-grade special effects materials by simply describing in natural language or drawing trajectories and leveraging the intelligent prompt word enhancement and automated generation capabilities. (2) Unlike the random videos output by general models, this invention enables the model to stably generate special effects videos with clean backgrounds, fixed perspectives, and standardized formats through domain-specific data training and optimization. Combined with the color filter rendering scheme, it fundamentally solves the common industry pain points of AI-generated special effects such as "difficulty in image cutout and edge noise", ensuring the high availability and standardization of the generated results, and enabling the generated content to be used directly as game assets. (3) By introducing context reference and end-to-end alignment mechanism between video frames, the technical gap of "standby / continuous" special effects that are difficult to produce in seamless loop playback in existing AI videos is successfully solved, realizing the real dynamic looping capability necessary for game special effects, and the generated sequence frames can be smoothly connected end to end. (4) Users can draw custom motion trajectories and combine them with style control to ensure that the generated special effects strictly comply with the logical constraints of game design while taking into account creativity and controllability. (5) Using VLM for semi-supervised annotation solves the problems of difficult and costly video data labeling, making it possible to rapidly iterate and train vertical domain models, build an efficient data production pipeline, and is particularly suitable as a game development engine plugin, cloud game material generation platform and AIGC art auxiliary tool. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a flowchart illustrating the model training and data processing during the implementation of this invention; Figure 3 This is a schematic diagram illustrating the principle of the first and last loop generation in this invention; Figure 4 This is a schematic diagram of the system structure of the present invention; Figure 5 The following are schematic diagrams illustrating the effects of the present invention, wherein (a) is the original image, (b) is a schematic diagram illustrating the traditional image cutout effect, and (c) is a schematic diagram illustrating the effect output after rendering by the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] This invention relates to an end-to-end automated generation and processing method for loopable game effects. By constructing a regularized game effects dataset and combining low-rank adaptive fine-tuning, semi-supervised annotation, and post-processing algorithms, it achieves end-to-end automation from text / trajectory input to usable game sequence frame output.

[0021] The method includes the following steps: (1) Construct a regularized game special effects video dataset, and adjust the parameters of the pre-trained video generation base model based on the dataset to obtain a vertical game special effects generation model. (2) Obtain the user input information, input the vertical game special effects generation model, and generate a game special effects video with a low brightness background. The low brightness background is a monochrome background and the brightness value is lower than a preset threshold. (3) Process the generated game effects video and output a sequence of frame images adapted for game engine rendering.

[0022] The method will be further explained below with reference to specific implementation methods.

[0023] (1) Construct a regularized game special effects video dataset, and adjust the parameters of the pre-trained video generation base model based on the dataset to obtain a vertical game special effects generation model. (1-1) Constructing a regularized game effects video dataset After acquiring the original special effects video footage and uniformly processing the background to black, the video footage is spatiotemporally normalized. The processed video footage is then classified, and data augmentation is performed within the same category to obtain a regularized game special effects video dataset.

[0024] Specifically, during player-versus-player combat, the release of skills requires synchronized attack actions and corresponding special effects. Only when the two are combined can there be a better display effect. Therefore, based on the visual feedback requirements of the game, special effect materials with clear "attack judgment", "status duration" and "hit feedback" logic are selected from the open source engine market, including but not limited to. The background pixels in the RGB channels of all materials are uniformly cleaned to pure black values ​​(R=0, G=0, B=0). This processing to black is to cooperate with the subsequent screen rendering technology and ensure that the AI ​​learns the feature distribution of "black is transparent". Spatiotemporal normalization refers to uniformly resampling video footage to a fixed frame rate (e.g., 30fps) and a fixed resolution (e.g., 512x512), and truncating or padding the duration to a standard length (e.g., 2 seconds) to ensure tensor alignment during model training. Subsequently, a classification and labeling system was established for different game perspectives such as side-scrolling 2D and top-down. Under the same type of special effect, such as "explosion", multiple material variations were created, including but not limited to pixel style, cartoon rendering, and realistic lighting effects, to ensure the model's generalization ability. The classification here can be configured by those skilled in the art according to actual needs.

[0025] (1-2) Semi-supervised automatic annotation In this invention, semi-supervised automatic annotation based on the Visual Language Model (VLM) is used to generate descriptive text for the regularized game effects video dataset, thereby achieving human-computer collaboration.

[0026] In the specific implementation process, in order to cope with the cold start, manual fine labeling is first performed. A small sample (e.g., 500) of extremely high-quality special effects videos are selected and manually labeled by experts to describe their features such as color, shape, dynamic patterns, and duration. Using the above manually labeled data, open-source visual language models, such as the Multimodal Visual Language Model (VLM), are fine-tuned to enable them to master the special prompts in the field of game special effects. The fine-tuned VLM is then used to perform batch inference on massive amounts of unlabeled data to generate high-precision descriptive text.

[0027] In practical applications, confidence filtering and manual verification can also be set up. Low-quality labels can be filtered out by setting a confidence threshold, and the sampled data can be manually checked. The corrected data can then be fed back into the model for iterative optimization, thereby greatly improving the efficiency of data preparation.

[0028] (1-3) Parameter adjustment and generation model of special effects for vertical games The parameters of the pre-trained video generation pedestal model are adjusted using a low-rank adaptive technique. This is achieved by injecting a trainable low-rank matrix into the attention layer of the pre-trained video generation pedestal model and freezing the original parameters of the pedestal model during training, while only updating the low-rank matrix parameters.

[0029] Specifically, in this embodiment, a diffusion transformer video model pre-trained on massive amounts of general data is used as the base model for video generation. Low-rank adaptation (LoRA) is employed to inject a low-rank matrix into the attention layers of the base model for training. By controlling the training parameters, the model is made to "overfit" to the physical laws of game effects (such as particle dissipation) and pure black background features, while retaining the image quality capabilities of the base model. This "overfitting" is actually domain adaptation, which forces the model to "forget" the complex background generation capabilities of general videos, making its weight distribution highly concentrated in the feature space of "pure black background + bright special effects subject", thereby ensuring that the generated video does not contain irrelevant background clutter and can be directly used as a game asset.

[0030] In practical applications, those skilled in the art can replace other available parameter adjustment techniques and corresponding tools as needed for parameter adjustment.

[0031] (2) Obtain the user input information, input the vertical game special effects generation model, and generate a game special effects video with a low brightness background. The low brightness background is a monochrome background and the brightness value is lower than a preset threshold. In this invention, the information input by the user includes a text description or a combination of a text description and motion trajectory information.

[0032] In this invention, the low-brightness background is a black (monochrome) background.

[0033] (2-1) Enhanced intelligent prompt words When generating the game effects video, the user's input is expanded into professional prompts including but not limited to materials, lighting effects, and dynamic details through a large language model; based on the semantics of the expanded prompts, a matching model is automatically selected from multiple vertical game effects generation models of different styles for generation.

[0034] Specifically, a large language model is introduced into the intermediate layer architecture of the agent as an intermediate layer for intent recognition and prompt word enhancement, which is used to enhance logic and route matching; Enhanced logic refers to the process where, when a user inputs a simple word (such as "fireball"), the LLM expands it into a professional prompt word that includes materials, lighting, and particle dynamics, based on a preset game art description template. Route matching refers to the system maintaining a keyword-model mapping table or using vector similarity retrieval. LLM analyzes the semantic embedding of user intent, calculates its distance with different style fine-tuning models (vertical models), such as pixel-style fine-tuning models and realistic-style fine-tuning models, and automatically activates the weight of the model with the highest matching degree for generation.

[0035] (2-2) Context-referenced loop video generation When generating the game effects video, the features of the first frame of the generated sequence are injected as constraints into the generation process of the last frame, or the end of the video sequence is filled in using video restoration technology with the first frame as a reference. The generated video has smooth frame transitions between its first and last frames, and plays in a loop.

[0036] Specifically, to address the requirement of continuous frame connections for the game's "idletime effect," a latent spatial constraint mechanism for the first and last frames is adopted. First, the first frame is anchored, i.e., the first frame (Frame 0) of the video is generated. When generating the last frame (Frame N), random sampling is not performed. Instead, the latent feature map of the first frame is used as a strong constraint condition. Alternatively, video restoration technology can be used to mask the tail area of ​​the video, and the first frame is used as a reference image to fill in the tail frame. The pixel distribution trend of the tail frame is forced to converge towards the first frame, thereby eliminating visual jumps (flickering) during loop playback and achieving pixel-level continuity.

[0037] (2-3) Generation based on trajectory control When the user input includes motion trajectory information, the motion trajectory information is converted into spatial control signals to guide the generated special effects subject to move along the specified trajectory when generating the game special effects video. In this invention, Wan-move or similar ControlNet technology is integrated. The front end allows users to draw motion paths, and the back end converts the paths into control signals (such as optical flow maps or position codes) to guide the model to follow a specified trajectory when generating pixels. This is a collaboration of multiple models and workflows. In the process, the models and workflows are integrated into a single intelligent agent. Specifically, the user provides a canvas where the user draws lines or shapes as motion paths. The server converts the user-drawn paths into spatial location codes or optical flow maps, and integrates spatial control networks, such as ControlNet or Adapter modules, to inject the aforementioned control signals into the cross-attention layer or spatial convolutional layer of the video generation model (base), guiding the generated main pixels to move strictly along preset coordinates, thereby achieving precise control over the trajectory, especially the "ballistics" and "skill range".

[0038] (3) Process the generated game effects video and output a sequence of frame images adapted for game engine rendering.

[0039] The generated game effects video is processed including color filtering rendering optimization and automated frame extraction.

[0040] (3-1) Color Screen Rendering Optimization Process Color filtering and rendering optimization processes generate videos with pure black backgrounds; In the game engine, the color blending mode is applied for rendering, making the pure black background completely transparent, while the bright parts of the special effects are superimposed on the brightness of the game background.

[0041] Since AI models often struggle to generate perfect transparency channels directly, resulting in white edges or unwanted colors, this invention employs "pure black background generation" combined with color filtering rendering optimization to address this issue. Specifically, the rendering of the color filter blending mode satisfies... Target = 1-(1-Source)*(1-Dest) Where Target is the final screen display color, Source is the main color of the effect after removing the background, and Dest is the original color of the game background. When the generated effect background is pure black, i.e. RGB=0, the above formula becomes Target= 1-(1-0)*(1-Dest)= Dest, meaning that black pixels do not affect the background at all and appear completely transparent. When the generated effect is a bright color, the formula calculation result will be brighter than the original background, thus simulating the visual effect of glowing and overlapping.

[0042] In this invention, a color filter shader is used for rendering in the game engine. By utilizing its mathematical property of "black is transparent but white is opaque", a perfect semi-transparent glowing effect can be achieved without an alpha channel. This avoids the jagged edges caused by image cutout. Furthermore, the RGB format saves 25% of video memory usage compared to the RGBA format, significantly reducing video memory consumption.

[0043] (3-2) Automated frame extraction processing In automated frame extraction, game effects videos are sampled into multiple static images at set time intervals and then arranged and stitched into a sequence of frame images. The arrangement is performed according to the number of rows and columns set by the user (e.g., 4x4).

[0044] The present invention also relates to an end-to-end automated generation and processing system for loopable game special effects, including a user terminal and a server terminal (C / S) connected by a communication interface.

[0045] (a) User terminal The client is used to receive information input by the user and send it to the server, and to receive and display the results generated by the server. Specifically, the user terminal is deployed on the user's local device, including but not limited to personal computers, or embedded in a plugin of the game engine, serving as the front-end interface for user interaction and a data transceiver terminal. Typically, the user terminal is equipped with an input receiving unit that provides a graphical user interface (GUI), including a text input box and a trajectory drawing canvas. The former is used for users to input simple natural language descriptions, such as "fireball," while the latter can provide drawing tools, allowing users to draw the motion path of special effects, such as parabolas and circular ranges, using a mouse or stylus. The user terminal is also configured with a data sending unit, which encapsulates the text description and / or trajectory coordinate data input by the user into a network request through API calls and sends it to the server; after receiving the sequence frame image returned by the server, it displays it in the user interface.

[0046] (ii) Server side The server side is typically deployed in the cloud or on a local server cluster, carrying the core AI models and processing logic.

[0047] The server side includes a prompt word processing module, a model routing and generation module, and a post-processing module. The model routing and generation module and the post-processing module execute the end-to-end loopable game effect automated generation and processing method.

[0048] The prompt word processing module is used to call large language models, such as Qwen, to enhance the received information; Specifically, upon receiving the original text description (such as "fireball") from the user, LLM automatically expands the simple words into professional prompts containing details such as materials (such as "flame"), lighting effects (such as "glowing"), and dynamics (such as "burst" and "excited state") based on the preset "game art description template," thus enhancing the original text description into text with specific details and dynamic features.

[0049] The model routing and generation module stores a pre-trained vertical game effects generation model, which is used to generate game effects videos with low-brightness backgrounds based on the enhanced information and the vertical game effects generation model. Specifically, storage space is configured to store multiple vertical game special effects generation models, which are obtained by fine-tuning general video diffusion models (such as Stable Video Diffusion and Wan2.1) using low-rank adaptive technology; The semantics of the enhanced professional prompts are analyzed by the routing matching unit, or the trajectory data input by the user is parsed. By calculating the similarity between the semantic embedding vector and the description vector of each model, or by matching the trajectory control requirements, one or more of the most suitable generative models are automatically selected. The selected model and its corresponding stylized model weights are loaded into the video generation unit. If the user inputs a motion trajectory, the trajectory data is converted into spatial control signals synchronously using a trajectory control network (such as Wan-move or ControlNet). During the generation process, if the user specifies that a loop effect is required, the context reference mechanism is activated, and the features of the first frame of the generated sequence are forcibly injected as a condition into the generation process of the last frame. Finally, output a dynamic effect video with a pure black (RGB 0,0,0) background that matches the user's intent.

[0050] The post-processing module is used to process the generated game effects video and output a sequence of frame images adapted for game engine rendering. Specifically, the system receives the video stream output by the model routing and generation module, samples it according to the user-preset or system default frame interval to obtain an image frame sequence, then arranges these image frames according to the specified number of rows and columns (such as 4*4) and splices them into a large sequence frame image (Sprite Sheet); converts the large sequence frame image into an image format common to game engines and packages it into a game effects resource package that can be directly imported.

[0051] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0052] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0053] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0054] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0055] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0056] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An end-to-end automated method for generating and processing loopable game special effects, characterized in that: A rule-based game effects video dataset is constructed, and the parameters of a pre-trained video generation base model are adjusted based on the dataset to obtain a vertical game effects generation model. Obtain user input information, input the vertical game special effects generation model, and generate a game special effects video with a low-brightness background. The low-brightness background is a monochrome background and the brightness value is lower than a preset threshold. The generated game effects video is processed to output a sequence of frame images adapted for game engine rendering.

2. The end-to-end automated generation and processing method for loopable game special effects according to claim 1, characterized in that: After acquiring the original special effects video footage and uniformly processing the background to black, the video footage is spatiotemporally normalized. The processed video footage is then classified, and data augmentation is performed within the same category to obtain a regularized game special effects video dataset.

3. The end-to-end automated generation and processing method for loopable game special effects according to claim 2, characterized in that: Descriptive text is generated from the regularized game effects video dataset using semi-supervised automatic annotation.

4. The end-to-end automated generation and processing method for loopable game special effects according to claim 1, characterized in that: The parameters of the pre-trained video generation pedestal model are adjusted using a low-rank adaptive technique. This is achieved by injecting a trainable low-rank matrix into the attention layer of the pre-trained video generation pedestal model and freezing the original parameters of the pedestal model during training, while only updating the low-rank matrix parameters.

5. The end-to-end automated generation and processing method for loopable game special effects according to claim 1, characterized in that: When generating the game effects video, the user's input is expanded into professional prompt words through a large language model; based on the semantics of the expanded prompt words, the matching model is automatically selected from multiple vertical game effects generation models of different styles for generation.

6. The end-to-end automated generation and processing method for loopable game special effects according to claim 5, characterized in that: When generating the game effects video, the features of the first frame of the generated sequence are injected as constraints into the generation process of the last frame, or the end of the video sequence is filled in using video restoration technology with the first frame as a reference. The generated video has smooth frame transitions between its first and last frames, and plays in a loop.

7. The end-to-end automated generation and processing method for loopable game special effects according to claim 1, characterized in that: When the user inputs information including motion trajectory information, the motion trajectory information is converted into spatial control signals when generating the game special effects video, guiding the generated special effects subject to move along the specified trajectory.

8. The end-to-end automated generation and processing method for loopable game special effects according to claim 1, characterized in that: The generated game effects video is processed including color filtering rendering optimization and automated frame extraction. Color filtering and rendering optimization processes generate videos with pure black backgrounds. Automated frame extraction processes sample game effects videos into multiple static images at set time intervals and arrange and stitch them together into a sequence of frame images.

9. The end-to-end automated generation and processing method for loopable game special effects according to claim 8, characterized in that: In the game engine, the color blending mode is applied for rendering, making the pure black background completely transparent, while the bright parts of the special effects are superimposed on the brightness of the game background.

10. An end-to-end automated generation and processing system for loopable game special effects, characterized in that: This includes the user terminal and server terminal connected via communication interfaces; The client is used to receive information input by the user and send it to the server, and to receive and display the results generated by the server. The server-side includes: The prompt word processing module is used to call the large language model to enhance the received information; The model routing and generation module stores a pre-trained vertical game effects generation model, which is used to generate game effects videos with low-brightness backgrounds based on the enhanced information and the vertical game effects generation model. The post-processing module is used to process the generated game effects video and output a sequence of frame images adapted for game engine rendering. The model routing and generation module and the post-processing module execute the end-to-end automated generation and processing method for loopable game effects as described in any one of claims 1 to 9.

Images