A branch generation system and method for game scenario interactive animation

Abstract

The application discloses a game plot interactive animation branch generation system and method, relates to the technical field of game interactive plot generation, and comprises a multi-modal interactive perception layer, a narrative kernel and a branch generation layer, a procedural animation generation layer and a real-time rendering presentation layer. The hierarchical branch is automatically triggered through a narrative entropy quantization model. The plot branch is dynamically generated based on the constraint basis of a world view knowledge graph, a character personality model and a narrative anchor point, and is verified for three-layer self-consistency, including causality consistency, character setting consistency and main line convergence. The procedural assembly generation of an animation segment is completed by relying on a prefabricated asset pool. Through the full-link cooperation of dynamic branch reasoning and procedural animation generation, the application supports the common triggering of branches by explicit operations and implicit behaviors, greatly reduces the content production and iteration cost, guarantees the self-consistency of the narrative logic and the controllability of the main line, and realizes the double improvement of interactive narrative immersion and production efficiency.

Description

Technical Field

[0001] This invention relates to the field of game interactive story generation technology, specifically to a branch generation system and method for game interactive animation. Background Technology

[0002] With the rapid development of digital content formats such as interactive narrative games and interactive anime series, story branching systems have become a core module for enhancing player engagement and immersion. Currently, mainstream interactive story branching solutions in the industry generally adopt a predefined branch tree architecture. The creative team pre-plans all plot developments and branch nodes, and produces corresponding animated clips, voice acting, and interactive resources. Players trigger different branch paths by clicking options at fixed nodes.

[0003] The existing branching storyline design has several inherent flaws: First, the number of branches is rigidly constrained by production costs, resulting in a prominent problem of "pseudo-choices." Each storyline branch requires its own script, animation, camera work, voice acting, and other resources. Production costs increase exponentially with the number of branches, leading to a very limited number of effective core branches in the final product. A large number of choices are merely decorative "pseudo-choices" that do not affect the main storyline. Player decisions cannot substantially change the plot development, which will weaken player trust in the long run and diminish the core value of interactive narrative.

[0004] Secondly, the branch triggering mechanism is rigid and inflexible, resulting in insufficient interactive immersion. The existing solution relies entirely on preset explicit pop-up options for branch triggering, which can only respond to explicit actions such as player clicks. It cannot recognize implicit behaviors such as player scene movement trajectory, item dwell time, and operation response rhythm, and it is also difficult to adapt to natural interactive inputs such as voice and emotional expression. The branch nodes are strongly disconnected from the plot flow, frequently interrupting the narrative continuity and failing to achieve a natural and smooth immersive interactive experience.

[0005] Third, the cost of content iteration and adjustment is high, and the development efficiency is low. The predefined branch tree is a strongly coupled structure. Any adjustment to the plot logic or character design may lead to simultaneous modifications to the scripts and animation resources of all subsequent branches. The adjustment chain is long and the rework cost is high. Content updates require the re-production of complete animation segments, making it difficult to quickly iterate and optimize based on player feedback, and it cannot adapt to the content update needs of long-term operation.

[0006] Some studies have attempted to introduce large language models to generate narrative text, but they generally lack systematic narrative constraints and verification mechanisms, and problems such as logical contradictions, character collapse, and loss of control of the main plot are prone to occur within the generated text. Moreover, most solutions only stay at the text generation stage and do not have a corresponding procedural animation generation chain, so they cannot achieve full-chain automation from plot branches to playable animations, making it difficult to directly implement them in actual game and interactive animation products. Summary of the Invention

[0007] The purpose of this invention is to provide a branch generation system and method for interactive animations in game storylines, in order to solve the problems of low freedom of game storylines and stiff interaction caused by the manual enumeration of branches in the prior art.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] According to the first aspect of this disclosure, a branch generation system for interactive animation of game plot is proposed, including a multimodal interaction perception layer, a narrative kernel and branch generation layer, a procedural animation generation layer and a real-time rendering and presentation layer;

[0010] The multimodal interaction perception layer is used to collect multi-dimensional interactive inputs from players, and to uniformly transform various inputs into structured interactive events with attribute tags and output them.

[0011] The narrative kernel and branch generation layer are used to receive the structured interactive events, dynamically generate and verify plot branches based on preset narrative constraint rules, and output standardized storyboard instructions.

[0012] The procedural animation generation layer is used to receive the standardized storyboard instructions and call the pre-made animation asset pool to complete the procedural assembly and generation of animation clips.

[0013] The real-time rendering layer is used to play generated animated story clips and present players with interactive entry points for corresponding story branches.

[0014] Furthermore, the narrative kernel and branch generation layer include a narrative ontology library, a narrative state machine, a dynamic branch generation module, a narrative self-consistency verification module, and a storyboard script generation module;

[0015] The narrative ontology library is used to store worldview knowledge graphs, character personality models, and preset narrative anchors, providing a rigid constraint basis for narrative content;

[0016] The narrative state machine is used to record the full-link narrative state data and causal event chain in real time, calculate the narrative entropy value based on the received structured interactive events, and output a branch generation instruction when the narrative entropy value reaches a preset threshold.

[0017] The dynamic branch generation module is used to receive the branch generation instruction, generate several candidate plot branches by combining the narrative constraints and the current narrative state, and select the effective branches by scoring and sorting.

[0018] The narrative self-consistency verification module is used to perform a three-level consistency verification on the selected candidate plot branches and output the qualified plot branches.

[0019] The storyboard generation module is used to convert the qualified plot branches into standardized storyboard instructions that include camera parameters, character actions, and dialogue content.

[0020] Furthermore, the input parameters for the narrative state machine to calculate the narrative entropy value include the weight of interactive events, the degree of deviation of events from the main storyline, and the distance of the current plot position from the next narrative anchor point; the narrative entropy value corresponds to three levels of branch levels, where low entropy value corresponds to automatically executed micro-branches that do not require player selection, medium entropy value corresponds to secondary branches that affect the mid-term plot direction, and high entropy value corresponds to core branches that affect the core main storyline ending.

[0021] Furthermore, the dynamic branch generation module is also equipped with a large language model that has been fine-tuned specifically for narrative script data. When generating candidate plot branches, it simultaneously outputs four types of elements: plot direction description, character reaction content, changes in world variables, and subsequent narrative impact. These elements are then scored and ranked according to three dimensions: narrative rationality, player preference matching degree, and plot tension.

[0022] Furthermore, the three-layer consistency check of the narrative self-consistency verification module specifically includes:

[0023] Causal consistency verification is performed by executing graph reasoning based on the worldview knowledge graph to verify the logical consistency between branch events and the chain of causal events that have already occurred.

[0024] Character consistency verification checks the degree of match between the words, actions, and decisions of the characters in the branches and the corresponding character personality models, and performs targeted corrections on mismatched content;

[0025] Main plot convergence check: Check whether the branching paths can converge to the next preset narrative anchor point in the subsequent plot.

[0026] Furthermore, the procedural animation generation layer includes a layered asset library module, a storyboard analysis and assembly module, and a pre-loading cache module;

[0027] The hierarchical asset library module is used to store three types of reusable animation assets: character-bound assets, scene component assets, and narrative shot templates.

[0028] The storyboard parsing and assembly module is used to parse standardized storyboard instructions, match corresponding animation assets, complete action transitions, lip-sync and facial expression adaptation, and camera movement choreography, and assemble and generate complete and playable animated storyboard clips.

[0029] The preloading cache module is used to preload the animation assets corresponding to high-probability branches in advance based on the sorting results of candidate plot branches, and cache the generated plot animation clips.

[0030] Furthermore, the multi-dimensional interactive inputs collected by the multimodal interaction perception layer include explicit UI selection interactions, implicit behavioral operations within the scene, and optional voice emotion inputs; the attribute tags attached to the structured interactive events include event weights, emotion tags, and behavior types.

[0031] Furthermore, the branch generation system also includes a data closed loop and iteration layer, which stores the player's full branch paths and interaction behavior data, optimizes the branch generation sorting rules and narrative entropy threshold parameters based on the full branch paths and interaction behavior data, and supports the addition of manually reviewed and calibrated content to the narrative ontology library.

[0032] According to the second aspect of this disclosure, a method for generating branching animations in game storylines is also proposed, comprising the following steps:

[0033] S1. Collect multi-dimensional interactive input from players and transform all types of input into structured interactive events with attached attribute tags;

[0034] S2. Calculate the narrative entropy value based on the current narrative state and the structured interactive events. When the narrative entropy value reaches a preset threshold, trigger the branch generation process.

[0035] S3. Based on the constraints of the preset worldview knowledge graph, character personality model and narrative anchor point, generate several candidate plot branches and complete the initial screening.

[0036] S4. Perform three-layer self-consistency checks on the candidate plot branches after the initial screening, namely causal consistency, character consistency and main plot convergence, to obtain qualified plot branches.

[0037] S5. Convert the qualified plot branches into standardized storyboard instructions, and call the pre-made animation asset pool to complete the procedural assembly and generation of plot animation segments.

[0038] S6. Play cutscenes and present players with interactive options for the corresponding branches. After receiving player selections, update the narrative status and advance the story.

[0039] Furthermore, in step S2, the narrative entropy value is determined by the interaction event weights. The degree to which the event deviates from the main plot And the remaining plot length from the current plot node to the next preset narrative anchor point. The result, obtained through joint calculation, is expressed as:

[0040]

[0041] in, This is a calibration coefficient used to match the overall numerical range of the narrative entropy value to a preset three-level branch trigger threshold; Indicates the remaining distance of the normalized anchor point. This refers to the current story segment, which is the total length between two adjacent narrative anchor points; The potential for divergence in the current plot is represented by the following: the further away from the next anchor point, the larger the remaining plot capacity, the more sufficient the buffer space for branch convergence, and the higher the narrative entropy value under the same interaction; the closer to the narrative anchor point, the more the plot tends to converge, and the lower the narrative entropy value under the same interaction.

[0042] Based on the calculated narrative entropy value When the branches are in different threshold ranges, they are divided into three levels: those below the first threshold are low-entropy micro-branches, which are used for automatic execution and do not pop up a selection entry; those between the first and second thresholds are medium-entropy secondary branches; and those above the second threshold are high-entropy core branches, which are used to pop up an interactive selection entry for the player.

[0043] Compared with existing technologies, this invention provides a branch generation system and method for interactive game story animations. By proposing a dynamic branch emergence mechanism under causal constraints, it uses a narrative entropy quantification model to drive hierarchical automatic branch triggering, complemented by a three-layer narrative self-consistency verification and narrative anchor point convergence system. Combined with a procedural animation generation pipeline, it achieves full-link dynamics. This expands story branches from finite paths of manual enumeration to near-infinite dynamic combinations, supporting both explicit operations and implicit behaviors to trigger branches, significantly enhancing interactive immersion and the realism of choices. Simultaneously, through pre-made asset reuse and procedural assembly, it greatly reduces the production and iteration costs of branch content, shortening the development cycle. Furthermore, relying on worldview knowledge graph constraints, dual causal and character verification, and anchor point convergence mechanisms, it achieves multiple improvements in narrative freedom, development efficiency, and immersive experience while ensuring stable and controllable narrative quality. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0045] Figure 1 A diagram illustrating the hierarchical relationships of the branch generation system for interactive game storylines provided by this invention.

[0046] Figure 2 This invention provides a diagram illustrating the relationship between the modules in the narrative kernel and branch generation layer.

[0047] Figure 3 A diagram illustrating the relationships between modules in the procedural animation generation layer provided by this invention;

[0048] Figure 4A flowchart illustrating the method for generating branching animations in interactive game storylines provided by this invention. Detailed Implementation

[0049] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0050] This embodiment uses the "Greystone Town" storyline from a fantasy role-playing game as an application scenario to elaborate on the complete operation process of the system and method.

[0051] After system startup, all module resources are loaded and initialization configurations are completed:

[0052] The narrative ontology library loads all narrative constraint resources for the corresponding plot segment: the worldview knowledge graph stores the entities and relationships of "Graystone Town" in a graph structure, including three major forces: the guard team, the merchant guild, and the slums; 12 NPC entities such as the guard captain Leon, the tavern owner Martha, and the homeless boy Cohen; as well as the social relationships, item function rules, and event causal rules between each entity; the character personality model configures structured parameters for each NPC, such as the homeless boy Cohen being set as quick-witted, highly vigilant, with survival as his core motivation, and wary of strangers; the guard captain Leon being set as upright and rigid, prioritizing duty, and having zero tolerance for violations; two core narrative anchor points are preset, the starting anchor point A is "the player arrives at the gate of Graystone Town and accepts the main quest to investigate the missing caravan", and the ending anchor point B is "the player goes to the city lord's mansion to report the progress of the investigation". The total plot length L0 between the two anchor points is set to 12 standard plot nodes, with each node corresponding to approximately 30 seconds of plot animation.

[0053] The procedural animation generation layer completes the loading of the asset library: The layered asset library contains three types of reusable assets. Character assets include standard rigging models of all characters, 27 sets of basic action clips (standing, walking, handing over objects, drawing swords, etc.), and 52 sets of facial expression fusion deformation targets. Scene assets include four types of modular scene components and corresponding interactive props, such as city gates, main streets, taverns, and slums. The shot template library has 18 sets of standardized shot scripts in four categories: dialogue shot-reverse shot, medium shot narrative, close-up emotional expression, and full shot environmental description.

[0054] Narrative state machine initialization global parameters: calibration coefficient K is initially set to 1.0, and the three-level branch trigger thresholds are the first threshold T1=0.2 and the second threshold T2=0.6 respectively.

[0055] Example 1: Scenario of Automatic Execution of Low-Entropy Micro-Branches

[0056] This embodiment corresponds to the unobtrusive micro-branch triggered by the player's implicit behavior, with no explicit selection pop-ups throughout.

[0057] 1. Interactive data collection and event structuring

[0058] After the player's character enters the main street of Graystone Town, they autonomously move the character to the location of the homeless boy Cohen, staying within 1.5 meters of him for 2.8 seconds, with the mouse cursor hovering over Cohen's character model multiple times. The multimodal interaction perception layer identifies and structures this continuous behavior, classifying it as an implicit behavior of "scene-active interaction," and generates a standardized structured interaction event with the following attribute labels: event weight W=0.35, behavior type=close-range observation, emotion label=curiosity neutral, triggering object=homeless boy Cohen.

[0059] 2. Narrative Entropy Calculation and Branch Trigger Determination

[0060] After receiving a structured interactive event, the narrative state machine retrieves the current global narrative state: 3 plot nodes have been completed, and the number of remaining plot nodes to the next anchor point B is L=9. The normalized anchor point remaining distance L / L0=9 / 12=0.75 is calculated. By comparing the path with the preset main line expected path (going straight along the main street to the tavern to inquire about information), the degree of deviation of the event from the main line is calculated to be D=0.4, which belongs to low deviation behavior.

[0061] Substitute the parameters into the narrative entropy calculation formula The calculated value is S = 1.0 × 0.35 × 0.4 × 0.75 = 0.105. Compared with the preset threshold, S = 0.105 < T1 = 0.2, which is determined to be a low-entropy micro-branch. The branch generation process is automatically triggered, and no explicit selection UI is displayed to the player.

[0062] 3. Candidate Branch Generation and Filtering

[0063] The dynamic branch generation module receives branch generation instructions, calls a large language model specifically fine-tuned with narrative script data, and takes into account the current narrative state, corresponding entity information in the worldview knowledge graph, Cohen's personality model, the causal event chain that has occurred, and the attributes of the current interaction event. The model generates 3 sets of candidate micro-branch schemes: ① Cohen glances at the player warily, turns around and hides in a nearby alley; ② Cohen actively approaches the player to beg for copper coins; ③ Cohen lowers his voice to remind the player that the town is not safe recently and to avoid going out at night.

[0064] The module quantifies and scores candidate branches based on three dimensions: narrative rationality, player preference matching, and plot tension. Ultimately, group ③ received the highest overall score and was determined as a valid branch, outputting the complete branch elements: the plot development is that the homeless boy takes the initiative to warn the player; the character's reaction is that Cohen speaks in a low voice and frequently scans his surroundings; the world variable changes are "Cohen's favorability +5" and "unlocked clue: strangers appear in the town at night"; the subsequent narrative impact is that Cohen's cooperation increases when the player takes the initiative to talk to him later.

[0065] 4. Three-layer self-consistency verification

[0066] The narrative consistency verification module performs three-layer consistency checks on valid branches: Causal consistency check: It retrieves the worldview knowledge graph for graph reasoning to confirm that the "strangers appearing at night" and the caravan's disappearance are logically consistent, and that the homeless boy's knowledge of street gossip aligns with information acquisition logic, with no causal conflicts; Character consistency check: It confirms that Cohen's proactive warning while remaining vigilant perfectly matches his "shrewd and wary" personality model, with no character inconsistencies; Main plot convergence check: It confirms that this branch only adds clues and favorability variables, without changing the player's core objective of reporting to the city lord's mansion, and can converge normally to narrative anchor point B. If all three checks pass, the branch is deemed qualified.

[0067] 5. Storyboard conversion and animation assembly generation

[0068] The storyboard generation module converts qualified branches into standardized storyboard instructions, containing three consecutive shots: Shot 1 is a medium shot with a fixed camera position, showing the player character standing on the street, with Cohen approaching from the side, lasting 2 seconds; Shot 2 is a close-up over-the-shoulder shot, showing Cohen's profile from the player's perspective, with Cohen speaking in a low voice and glancing around, corresponding to the line "Don't wander around the streets at night lately, there are strange people around," lasting 3 seconds; Shot 3 is a medium shot following shot, showing Cohen lowering his head and walking quickly away, disappearing into the alleyway after speaking, lasting 2 seconds. Each instruction includes five types of parameters: character movement, facial expression weight, dialogue content, camera movement, and emotional tag.

[0069] The procedural animation generation layer receives storyboard instructions. After the storyboard parsing and assembly module parses the parameters, it matches the corresponding assets from the layered asset library: character actions call three sets of action segments: "standing and turning sideways," "speaking in a low voice," and "walking away quickly," achieving natural transitions between segments through animation state machines and motion blending technology; the facial expression system matches "alertness" with a weight of 0.7 and "tension" with a weight of 0.3; the lip-syncing module automatically generates corresponding lip-sync animations based on the dialogue text; the camera system calls two sets of preset templates: "medium shot fixed" and "close-up over the shoulder," automatically matching camera position parameters and camera movement rhythm. After assembly, a complete 7-second animated clip is generated, and the pre-loading cache module stores the clip in the local cache.

[0070] 6. Rendering and Status Updates

[0071] The real-time rendering layer seamlessly connects to the main storyline, automatically playing the micro-branch animation clips without any pop-ups interrupting the player's operation rhythm. After the animation ends, the narrative state machine synchronously updates the world variables, causal event chains, and current story node counts, allowing the story to continue naturally along the main storyline.

[0072] Example 2: Player Selection Scene in Medium Entropy Secondary Branch

[0073] This embodiment corresponds to the secondary branch triggered by the player's moderate deviation from the main storyline, and provides players with immersive options for selection.

[0074] When the story progresses to the 6th plot point, the player controls the character to walk to the guard checkpoint, actively draws the weapon, and continues to move in the direction of the guard.

[0075] 1. Interactive data collection and event structuring

[0076] The multimodal interaction perception layer collects the continuous operations of "drawing the weapon" and "moving towards the guard" and identifies them as high-weight explicit hostile tendencies. It generates a structured interaction event: event weight W=0.85, behavior type=hostile tendencies operation, emotion label=aggressive, triggering object=guard captain Leon.

[0077] 2. Narrative Entropy Calculation and Branch Trigger Determination

[0078] The narrative state machine retrieves the current state: remaining plot nodes L=6, normalized anchor point remaining distance L / L0=6 / 12=0.5; event deviation from the main storyline D=0.85, severely deviating from the expected behavior of "peaceful passage". Substituting into the formula, we get S=1.0×0.85×0.85×0.5≈0.361. Comparing with the threshold, we get T1=0.2<S<T2=0.6, which is determined to be a medium-entropy secondary branch, triggering the explicit branch generation process, requiring an interactive selection entry point to be presented to the player.

[0079] 3. Candidate Branch Generation and Filtering

[0080] The dynamic branch generation module generates four candidate branch schemes, which are then scored and sorted to select three valid branches: ① Forcefully pass through, engage in battle with the guards, defeat them to pass, triggering a city-wide wanted status for the guards; ② Show the previously obtained merchant guild token, explain your purpose and request passage, Leon is skeptical but grants passage; ③ Put away your weapon, turn around and leave, detouring through the slums' secret passages to reach your destination, triggering additional exploration events. Each branch contains four complete elements: plot development, character reactions, changes in world variables, and subsequent narrative impact.

[0081] 4. Three-layer self-consistency verification

[0082] The narrative self-consistency verification module performed verification group by group: Group ①, the player drawing the sword and causing the conflict is consistent with causal logic, the guards issuing a wanted notice is consistent with the faction rules, and Leon's resolute reaction to stop them is consistent with his upright character. The player can still go to the city lord's mansion after clearing the level, only changing the guards' attitude afterward. It can converge normally to anchor point B, and the verification passed; Group ②, the merchant guild token has a valid pass in the world view, and Leon's skeptical attitude is consistent with his character. It can converge the main storyline, and the verification passed; Group ③, the secret passage in the slums exists in the world view knowledge graph, and the detour behavior is logical. It can converge the main storyline, and the verification passed. All three groups of branches were judged to be qualified.

[0083] 5. Storyboard generation and asset preloading

[0084] The storyboard generation module generates standardized storyboard instructions for each of the three grid branches, which are then synchronously passed to the procedural animation generation layer. The preloading cache module prioritizes the preloading of all animation assets corresponding to the second group of branches, which is ranked first, based on the branch sorting priority. The other two groups are preloaded asynchronously in the background to ensure that the player can play the selected content quickly.

[0085] 6. Interactive presentation and plot progression

[0086] The real-time rendering layer presents an immersive selection interface within the scene. Three options float in the corresponding positions of the scene as interactive icons, replacing traditional pop-up options. After the player selects the second branch, the system immediately plays the corresponding animation clip showing the equipment completed; the narrative state machine synchronously updates the world variables: "Guardian Team Favorability -10" and "Merchant Guild Power Connection +15", recording the corresponding causal events, and the plot continues to advance.

[0087] Example 3: Scenarios Influencing the High-Entropy Core Branch Mainline

[0088] This example corresponds to a high-level branch triggered by the player's core decision, which will directly change the form of the narrative anchor point and the subsequent main storyline.

[0089] As the story progresses to the fourth plot point, players obtain crucial evidence of the missing caravan during their exploration, discovering that the city lord is colluding with a smuggling ring. Players then choose to "destroy the evidence and accept the city lord's bribe."

[0090] 1. Interactive data collection and event structuring

[0091] The multimodal interaction perception layer identifies the operation as a core plot decision-making interaction and generates a structured interaction event: event weight W=0.95, behavior type=core plot decision, emotion tag=interest tendency, triggering object=main core plot.

[0092] 2. Narrative Entropy Calculation and Branch Trigger Determination

[0093] Narrative state machine calculation parameters: Remaining plot nodes L=8, normalized anchor point remaining distance L / L0=8 / 12≈0.667; event deviation from the main plot D=0.95, highly deviating from the expected direction of the main plot of "revealing the truth". Substituting into the formula, we get S=1.0×0.95×0.95×0.667≈0.603. Comparing with the threshold, we get S>T2=0.6, which is judged as a high-entropy core branch, which will directly affect the final form of anchor point B and the subsequent main plot ending.

[0094] 3. Candidate Branch Generation and Filtering

[0095] The dynamic branch generation module generates three core branch directions, which are then presented to the player after selection: ① Expose the truth, hand over the evidence to the guard team, trigger the city lord's counterattack plot, and head towards the righteous ending line; ② Destroy the evidence and accept bribes, help the city lord conceal the truth, and head towards the dark ending line; ③ Anonymously report the evidence, secretly promote the investigation, and head towards the neutral and secluded ending line.

[0096] 4. Three-layer self-consistency verification

[0097] The narrative self-consistency verification module verified each group individually: At the causal level, all three choices conformed to the power struggle logic within the worldview, with no logical contradictions; at the character level, the corresponding NPCs' reactions matched their personality models; at the main plot convergence level, the three branches corresponded to three different forms of anchor point B (the city lord is controlled, the city lord continues to hold power, and the city lord is secretly investigated), all falling within the acceptable variations of the narrative anchor point and satisfying the main plot convergence requirements. All three branches passed the verification.

[0098] 5. Select execution and mainline convergence

[0099] After the player makes a choice, the system plays the corresponding animation clip, the narrative state machine updates the core world variables and causal event chain, and the plot advances to the corresponding narrative anchor point B. The subsequent main storyline will be permanently changed based on this choice.

[0100] Data closed-loop iterative implementation

[0101] Throughout system operation, the data loop and iteration layer collect all player data in real time, including metrics such as selection rate, dwell time, reselection rate, and story completion rate for each branch. At fixed operational intervals, the system optimizes parameters based on accumulated data: for example, if statistics show that the dwell trigger rate for wandering NPC scenes is too high, leading to excessively frequent micro-branch triggers, the event weight W for this type of interaction is lowered, or the first threshold T1 is raised to optimize narrative entropy trigger sensitivity; conversely, if statistics show that the player selection rate for a certain type of branch is extremely low and feedback is poor, the scoring weight of the dynamic branch generation module is adjusted, reducing the ranking priority of similar branches. Simultaneously, the manual review team regularly extracts high-popularity branch content for review, adding high-quality branch storylines and character dialogue to the narrative ontology sample library for subsequent fine-tuning and iteration of the large narrative model, continuously improving the quality and stability of generated content.

[0102] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A branch generation system for interactive animations in a game's storyline, characterized in that, It includes a multimodal interaction perception layer, a narrative kernel and branch generation layer, a procedural animation generation layer, and a real-time rendering and presentation layer; The multimodal interaction perception layer is used to collect multi-dimensional interactive inputs from players, and to uniformly transform various inputs into structured interactive events with attribute tags and output them. The narrative kernel and branch generation layer are used to receive the structured interactive events, dynamically generate and verify plot branches based on preset narrative constraint rules, and output standardized storyboard instructions. The procedural animation generation layer is used to receive the standardized storyboard instructions and call the pre-made animation asset pool to complete the procedural assembly and generation of animation clips. The real-time rendering layer is used to play generated animated story clips and present players with interactive entry points for corresponding story branches.

2. The branch generation system for interactive game story animation according to claim 1, characterized in that, The narrative kernel and branch generation layer include a narrative ontology library, a narrative state machine, a dynamic branch generation module, a narrative self-consistency verification module, and a storyboard script generation module; The narrative ontology library is used to store worldview knowledge graphs, character personality models, and preset narrative anchors, providing a rigid constraint basis for narrative content; The narrative state machine is used to record the full-link narrative state data and causal event chain in real time, calculate the narrative entropy value based on the received structured interactive events, and output a branch generation instruction when the narrative entropy value reaches a preset threshold. The dynamic branch generation module is used to receive the branch generation instruction, generate several candidate plot branches by combining the narrative constraints and the current narrative state, and select the effective branches by scoring and sorting. The narrative self-consistency verification module is used to perform a three-level consistency verification on the selected candidate plot branches and output the qualified plot branches. The storyboard generation module is used to convert the qualified plot branches into standardized storyboard instructions that include camera parameters, character actions, and dialogue content.

3. The branch generation system for interactive game story animation according to claim 2, characterized in that, The input parameters for calculating the narrative entropy value of the narrative state machine include the weight of interactive events, the degree of deviation of events from the main storyline, and the distance of the current plot position from the next narrative anchor point. The narrative entropy value corresponds to three levels of branch levels, where low entropy value corresponds to automatically executed micro-branches that do not require player selection, medium entropy value corresponds to secondary branches that affect the mid-term plot direction, and high entropy value corresponds to core branches that affect the core main storyline ending.

4. The branch generation system for interactive game story animation according to claim 2, characterized in that, The dynamic branch generation module is also equipped with a large language model that has been fine-tuned with specific narrative script data. When generating candidate plot branches, it simultaneously outputs four types of elements: plot direction description, character reaction content, changes in world variables, and subsequent narrative impact. It is then scored and ranked according to three dimensions: narrative rationality, player preference matching degree, and plot tension.

5. The branch generation system for interactive game story animation according to claim 2, characterized in that, The three-layer consistency check of the narrative self-consistency verification module specifically includes: Causal consistency verification is performed by executing graph reasoning based on the worldview knowledge graph to verify the logical consistency between branch events and the chain of causal events that have already occurred. Character consistency verification checks the degree of match between the words, actions, and decisions of the characters in the branches and the corresponding character personality models, and performs targeted corrections on mismatched content; Main plot convergence check: Check whether the branching paths can converge to the next preset narrative anchor point in the subsequent plot.

6. The branch generation system for interactive game story animation according to claim 1, characterized in that, The procedural animation generation layer includes a layered asset library module, a storyboard analysis and assembly module, and a pre-loading cache module. The hierarchical asset library module is used to store three types of reusable animation assets: character-bound assets, scene component assets, and narrative shot templates. The storyboard parsing and assembly module is used to parse standardized storyboard instructions, match corresponding animation assets, complete action transitions, lip-sync and facial expression adaptation, and camera movement choreography, and assemble and generate complete and playable animated storyboard clips. The preloading cache module is used to preload the animation assets corresponding to high-probability branches in advance based on the sorting results of candidate plot branches, and cache the generated plot animation clips.

7. The branch generation system for interactive game story animation according to claim 1, characterized in that, The multi-dimensional interactive inputs collected by the multimodal interaction perception layer include explicit UI selection interactions, implicit behavioral operations within the scene, and optional voice emotion inputs; the attribute tags attached to the structured interactive events include event weight, emotion tags, and behavior types.

8. The branch generation system for interactive game story animation according to claim 1, characterized in that, The branch generation system also includes a data closed loop and iteration layer, which stores all branch paths and interaction behavior data of players, optimizes the branch generation sorting rules and narrative entropy threshold parameters based on the all branch paths and interaction behavior data, and supports the addition of manually reviewed and marked content to the narrative ontology library.

9. A method for generating branching animations in a game's storyline, characterized in that, Includes the following steps: S1. Collect multi-dimensional interactive input from players and transform all types of input into structured interactive events with attached attribute tags; S2. Calculate the narrative entropy value based on the current narrative state and the structured interactive events. When the narrative entropy value reaches a preset threshold, trigger the branch generation process. S3. Based on the constraints of the preset worldview knowledge graph, character personality model and narrative anchor point, generate several candidate plot branches and complete the initial screening. S4. Perform three-layer self-consistency checks on the candidate plot branches after the initial screening, namely causal consistency, character consistency and main plot convergence, to obtain qualified plot branches. S5. Convert the qualified plot branches into standardized storyboard instructions, and call the pre-made animation asset pool to complete the procedural assembly and generation of plot animation segments. S6. Play cutscenes and present players with interactive options for the corresponding branches. After receiving player selections, update the narrative status and advance the story.

10. The method for generating branching animations in a game's storyline according to claim 9, characterized in that, In step S2, the narrative entropy value is determined by the interaction event weights. The degree to which the event deviates from the main plot And the remaining plot length from the current plot node to the next preset narrative anchor point. The result, obtained through joint calculation, is expressed as: in, This is a calibration coefficient used to match the overall numerical range of the narrative entropy value to a preset three-level branch trigger threshold; Indicates the remaining distance of the normalized anchor point. This refers to the current story segment, which is the total length between two adjacent narrative anchor points; The potential for divergence in the current plot is represented by the following: the further away from the next anchor point, the larger the remaining plot capacity, the more sufficient the buffer space for branch convergence, and the higher the narrative entropy value under the same interaction; the closer to the narrative anchor point, the more the plot tends to converge, and the lower the narrative entropy value under the same interaction. Based on the calculated narrative entropy value When the branches are in different threshold ranges, they are divided into three levels: those below the first threshold are low-entropy micro-branches, which are used for automatic execution and do not pop up a selection entry; those between the first and second thresholds are medium-entropy secondary branches; and those above the second threshold are high-entropy core branches, which are used to pop up an interactive selection entry for the player.

Images