Map interaction-based real-time generation method and device for digital human marketing video of a merchant
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING FENGPING INTELLIGENT TECHNOLOGY CO LTD
- Filing Date
- 2026-02-25
- Publication Date
- 2026-05-29
Smart Images
Figure CN122120569A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of video processing technology, and specifically relates to a method and apparatus for real-time generation of digital human marketing videos for merchants based on map interaction. Background Technology
[0002] With the deep integration of digital map services and local life platforms, such as the integration of Gaode Maps and Meituan, and Baidu Maps and Dianping, precision marketing based on Geographic Information Systems (GIS) has become an important means for offline businesses to attract customers. Currently, mainstream map applications allow businesses to display basic information by marking Points of Interest (POIs) information, such as location, name, and category, and uploading text and images. When users search or browse maps, they can view this static information or be redirected to third-party platforms for more details.
[0003] Existing technical solutions typically fall into two categories:
[0004] First, the static information display mode. Merchants pre-create promotional images or coupon information and configure them at map POIs or in the backend system. When a user clicks, the system directly displays this pre-made content. The drawback of this mode is that the content is static, lacks dynamism and appeal, and is difficult to effectively capture user interest in a short period of time. At the same time, the content production cost is high, requiring professional graphic designers and planners, making it difficult to achieve large-scale, personalized coverage.
[0005] Second, there's the third-party link redirection model. Map service providers integrate with local service platforms via API interfaces. When a user clicks on a Point of Interest (POI), the map application redirects to the corresponding business page, such as Dianping (a Chinese review platform). While this model provides richer information, it disrupts the user's immersive experience within the map, resulting in a longer operation path and impacting conversion efficiency. Furthermore, the displayed content remains pre-existing and not generated in real time.
[0006] There is an urgent need in the field for a solution that can deeply integrate map interaction and AI generation technologies to achieve low-cost, high-efficiency, and high-real-time personalized marketing video automatic generation and distribution. Summary of the Invention
[0007] To address at least one of the aforementioned technical problems, this application provides a method and apparatus for real-time generation of digital human marketing videos for merchants based on map interaction.
[0008] The first aspect of this application is a method for real-time generation of digital human marketing videos for merchants based on map interaction, which mainly includes:
[0009] S1. Responding to user interaction on the map interface, capture the selected merchant's POI-ID;
[0010] S2. Based on the merchant's POI-ID, acquire and aggregate the merchant's multi-source data in real time. The multi-source data includes at least basic information, marketing activity data, and social media comment data, and construct a structured merchant feature vector.
[0011] S3. Based on the merchant feature vector, a large language model dynamically generates short video marketing copy that matches the merchant.
[0012] S4. Based on the marketing copy, generate corresponding text, audio stream, and subtitle timeline triplet data simultaneously;
[0013] S5. Based on the category information in the merchant feature vector, automatically match and call the corresponding digital human image from the digital human resource library;
[0014] S6. Using the voice stream, the selected digital human's lip movements are synchronized through phoneme mapping technology, and the preset action template of the digital human is triggered based on the keywords in the marketing copy.
[0015] S7. At the edge computing node, the driven digital human video stream is rendered and synthesized in real time. The synthesis includes at least: background replacement, overlaying dynamic subtitles corresponding to the subtitle timeline, and adding background music, and finally generating a complete marketing video file.
[0016] S8. The generated marketing video file is pushed back to the map interface for playback with low latency via a progressive transmission protocol.
[0017] Preferably, in step S2, the real-time acquisition and aggregation of multi-source data is achieved by calling one or more of the following APIs: map service provider API, Dianping API, and merchant's own CRM system API.
[0018] Preferably, in step S2, the constructed structured merchant feature vector is represented as follows:
[0019] V=[a,V_basic,b,V_marketing,c,V_social];
[0020] Among them, V_basic is the basic information sub-vector with weight a, V_marketing is the marketing sub-vector with weight b, and V_social is the social sub-vector with weight c. The weights a, b, and c are obtained by looking up a table based on the current time and the merchant category.
[0021] Preferably, in step S3, the dynamic generation of short video marketing copy is generated by a large language model based on a preset marketing script style template and the merchant feature vector, producing copy text with a video length of no more than 15 seconds.
[0022] Preferably, step S3 further includes:
[0023] Step S31: Dynamically construct the prompt word P based on the merchant feature vector V. The construction formula is as follows:
[0024] P=Template[Keywords(V)]+Constraints;
[0025] Wherein, Template is a preset marketing script style template; Keywords(V) is a function that extracts core keywords from feature vector V; Constraints are the generation constraints, including at least video duration constraints and sentence structure constraints;
[0026] Step S32: Input the prompt word P into the large language model to generate short video marketing copy.
[0027] Preferably, in step S6, the keyword-triggered preset action template includes:
[0028] A keyword-action mapping library is constructed. When a keyword from the mapping library is detected in a marketing copy, the digital human image is automatically triggered to perform the corresponding action, which includes gestures, facial expressions, or body animations.
[0029] Preferably, in step S6, when keyword K is identified, the probability P(A|K) of triggering the corresponding action A is calculated by the following formula:
[0030] P(A|K)=B(K)+η*Score(S);
[0031] Where B(K) is the preset baseline trigger probability of keyword K in the action mapping library, Score(S) is the overall sentiment score of the current marketing copy S, and η is the sentiment influence factor coefficient.
[0032] Preferably, in step S7, background replacement includes:
[0033] Based on the category information in the merchant feature vector, matching background materials are dynamically loaded from the background template library.
[0034] A second aspect of this application discloses a real-time generation device for merchant digital human marketing videos based on map interaction, mainly comprising:
[0035] The interactive data acquisition module is used to respond to user interactions on the map interface and capture the selected merchant's POI-ID.
[0036] The merchant feature vector construction module is used to acquire and aggregate multi-source data of the merchant in real time based on the merchant's POI-ID. The multi-source data includes at least basic information, marketing activity data and social media comment data, and constructs a structured merchant feature vector.
[0037] The short video marketing copy generation module is used to dynamically generate short video marketing copy that matches the merchant based on the merchant's feature vector using a large language model.
[0038] The triplet data extraction module is used to simultaneously generate corresponding text, audio stream, and subtitle timeline triplet data based on the marketing copy.
[0039] The digital human matching module is used to automatically match and call the corresponding digital human image from the digital human resource library based on the category information in the merchant feature vector;
[0040] The digital human parameter adjustment module is used to use the voice stream to drive the lip movements of the selected digital human image to synchronize through phoneme mapping technology, and to trigger the preset action template of the digital human based on the keywords in the marketing copy.
[0041] The video rendering module is used to perform real-time rendering and compositing of the driven digital human video stream at the edge computing node. The compositing includes at least: background replacement, overlaying dynamic subtitles corresponding to the subtitle timeline, and adding background music, and finally generating a complete marketing video file.
[0042] The video loading module is used to push the generated marketing video file back to the map interface for playback with low latency via a progressive transmission protocol.
[0043] A third aspect of this application is a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method for real-time generation of map-based interactive digital human marketing videos for merchants as described above.
[0044] This application achieves ultimate real-time performance and an immersive user experience, significantly reducing the production cost and barriers to personalized marketing content, and enhancing the dynamic relevance and appeal of marketing content. Attached Figure Description
[0045] Figure 1 This is a flowchart of a preferred embodiment of the map-based interactive method for real-time generation of digital human marketing videos for merchants. Detailed Implementation
[0046] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be described in more detail below with reference to the accompanying drawings. In the drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The described embodiments are only some, not all, of the embodiments of this application. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. The embodiments of this application will be described in detail below with reference to the accompanying drawings.
[0047] The first aspect of this application provides a method for detecting digital human video live streaming, such as... Figure 1 As shown, it mainly includes:
[0048] S1. Responding to user interaction on the map interface, capture the selected merchant's POI-ID;
[0049] S2. Based on the merchant's POI-ID, acquire and aggregate the merchant's multi-source data in real time. The multi-source data includes at least basic information, marketing activity data, and social media comment data, and construct a structured merchant feature vector.
[0050] S3. Based on the merchant feature vector, a large language model dynamically generates short video marketing copy that matches the merchant.
[0051] S4. Based on the marketing copy, generate corresponding text, audio stream, and subtitle timeline triplet data simultaneously;
[0052] S5. Based on the category information in the merchant feature vector, automatically match and call the corresponding digital human image from the digital human resource library;
[0053] S6. Using the voice stream, the selected digital human's lip movements are synchronized through phoneme mapping technology, and the preset action template of the digital human is triggered based on the keywords in the marketing copy.
[0054] S7. At the edge computing node, the driven digital human video stream is rendered and synthesized in real time. The synthesis includes at least: background replacement, overlaying dynamic subtitles corresponding to the subtitle timeline, and adding background music, and finally generating a complete marketing video file.
[0055] S8. The generated marketing video file is pushed back to the map interface for playback with low latency via a progressive transmission protocol.
[0056] In step S1, this application first interacts with the user to receive the merchant selected by the user. In step S1, the merchant POI-ID is a unique identification code assigned by the map service provider or platform (such as Gaode Map, Baidu Map, Meituan, Dianping, etc.) to each geographical location (Point of Interest) in its database. It is not only a merchant, but any location with geographical location information can be a POI, such as: restaurant, hotel, shopping mall, gas station, school, park, bus stop, building, or even a specific door address. Its core function is accurate positioning and data association.
[0057] After selecting a merchant, the system needs to collect data in the background, that is, to aggregate multi-source data in step S2.
[0058] In some alternative implementations, in step S2, the real-time acquisition and aggregation of multi-source data is achieved by calling one or more of the following APIs: map service provider API, Dianping API, and merchant's own CRM system API.
[0059] In some alternative implementations, the structured merchant feature vector constructed in step S2 is represented as follows:
[0060] V=[a,V_basic,b,V_marketing,c,V_social];
[0061] Among them, V_basic is the basic information sub-vector with weight a, V_marketing is the marketing sub-vector with weight b, and V_social is the social sub-vector with weight c. The weights a, b, and c are obtained by looking up a table based on the current time and the merchant category.
[0062] For example, suppose a user clicks on a restaurant called "La Fu" (Category: Sichuan Hot Pot). Here, V_basic might be [Category: Sichuan Hot Pot; Average Price: 150 RMB; Rating: 4.8 stars]; V_marketing might be [Current Discount: Yes; Discount Amount: 100 RMB off for purchases over 300 RMB]; and V_social is a quantifiable value, calculated, for example, by extracting keywords and sentiment values from the latest 100 reviews using a sentiment analysis model.
[0063] Positive keywords: ["Fresh tripe": 0.9; "Good service": 0.85; "Fragrant broth": 0.95] (Values represent emotional intensity).
[0064] Negative keyword: ["Long queues": -0.7].
[0065] After comprehensive calculation, V_social can be quantified as a high sentiment score, such as 0.88 (score range -1 to 1).
[0066] Weights a, b, and c can be found in the table below.
[0067]
[0068] For example, the system detects that "La Fu" has high real-time popularity at 7 pm on Saturday, and automatically reorganizes the weights according to the table above (a=0.2, b=0.3, c=0.5).
[0069] The final vector calculation is as follows:
[0070] V=[0.2,[Hot pot, 150 yuan, 4.8 stars],0.3,[Yes, 100 yuan off for every 300 yuan spent],0.5,[0.88]].
[0071] This merchant feature vector clearly indicates that when generating copy, 50% of the weight should be given to content from positive social media reviews. The above formula allows the system to move beyond mechanically listing data and intelligently determine "what should be said to the consumer at this moment to be most effective," demonstrating true intelligence.
[0072] In some optional implementations, in step S3, the dynamic generation of short video marketing copy is achieved by a large language model generating copy text with a video duration of no more than 15 seconds based on a preset marketing script style template and the merchant feature vector.
[0073] In some alternative implementations, step S3 further includes:
[0074] Step S31: Dynamically construct the prompt word P based on the merchant feature vector V. The construction formula is as follows:
[0075] P=Template[Keywords(V)]+Constraints;
[0076] Wherein, Template is a preset marketing script style template; Keywords(V) is a function that extracts core keywords from feature vector V; Constraints are the generation constraints, including at least video duration constraints and sentence structure constraints;
[0077] Step S32: Input the prompt word P into the large language model to generate short video marketing copy.
[0078] Following the above embodiments, in step S2, the feature vector V calculated for "La Fu" has been obtained, the core of which is "high reputation" and "discounts".
[0079] In step S31, the most prominent keywords are first extracted using Keywords(V). Because the weight c=0.5 is very high, this function will prioritize extracting high sentiment score keywords from the social subvectors: "fresh tripe" and "fragrant hot pot base". At the same time, the marketing keyword "spend 300 and get 100 off" will also be extracted.
[0080] Then, keywords are populated based on the style template provided by the template. For example, the template might be: "This is a [business category], and its main feature is [featured tags]. We currently have a [marketing campaign], come and experience it!"
[0081] Finally, add constraints as conditions, such as "Generate a Chinese marketing script of no more than 15 seconds, which should be colloquial, enthusiastic, and must contain the extracted keywords."
[0082] The above parameters are combined into a complete prompt word / statement and sent to the large model. For the above embodiment, the final generated statement is: "This is a Sichuan hot pot restaurant, its main features are fresh tripe and fragrant broth. We currently have a promotion of 100 RMB off for every 300 RMB spent, come and experience it! Based on the above information, please generate a short Chinese marketing script of no more than 15 seconds, requiring a conversational and enthusiastic tone." In step S32, the above statement is sent to the large language model, and the generated short video marketing script is, for example: "Hey, foodies, listen up! The beef tallow broth at La Fu Hot Pot is so fragrant! The signature tripe is fresh and crisp, and there's a super value offer of 100 RMB off for every 300 RMB spent. Let's make plans for a weekend get-together!"
[0083] Then, in step S4, the marketing copy is represented by triples for later retrieval.
[0084] Step S5 is used to match digital human figures; for example, catering businesses can choose chef costume images, etc.
[0085] Step S6 is used to adjust the lip movements and body movements of the selected digital human. Lip control technology is relatively mature, while body movements need to match the content of the current text and the type of business. Specifically, this application can trigger corresponding body movements through keywords, such as activating a thumbs-up gesture animation when "discount" is mentioned.
[0086] In some optional implementations, step S6, triggering the preset action template based on keywords, includes:
[0087] A keyword-action mapping library is constructed. When a keyword from the mapping library is detected in a marketing copy, the digital human image is automatically triggered to perform the corresponding action, which includes gestures, facial expressions, or body animations.
[0088] In some optional implementations, in step S6, when keyword K is identified, the probability P(A|K) of triggering the corresponding action A is calculated by the following formula:
[0089] P(A|K)=B(K)+η*Score(S);
[0090] Where B(K) is the preset baseline trigger probability of keyword K in the action mapping library, Score(S) is the overall sentiment score of the current marketing copy S, and η is the sentiment influence factor coefficient.
[0091] Continuing from the previous example, the large speech model LLM generated the following text: "Hey, foodies, listen up! The beef tallow broth at La Fu Hot Pot is so fragrant! The signature tripe is fresh and crisp, and there's a great deal of 100 off for every 300 spent. Let's make plans for a weekend get-together!"
[0092] In step S6, regarding the selection of body movements, firstly, a baseline trigger probability is assigned to the keyword K using B(K). For example, the preset B(discount) = 0.8, meaning that mentioning discounts has an 80% probability of triggering a thumbs-up gesture. Then, the emotional score of the text S is calculated using η*Score(S). For example, the preset emotional influence factor η = 0.2. In the above embodiment, the text S represents very positive and excited feelings, and the calculated Score(S) = 0.9. Therefore, the probability P(A|K) = P(thumbs-up gesture|"discount") = 0.8 + (0.2*0.9) = 0.98.
[0093] The system then typically generates a random number between 0.8 and 1, such as 0.85. Since 0.85 < 0.98, a "like" gesture is triggered. If the random number is 0.99, then because 0.99 > 0.98, the gesture is not triggered this time. This formula makes the digital human's behavior no longer mechanical and predictable (always a "like" when a discount is mentioned), but rather possesses a human-like uncertainty. The higher the overall emotional tone of the text, the greater the likelihood of triggering an action, and the more excited the human appears. This greatly enhances the vividness and realism of the digital human, avoiding the rigid experience of a "robot."
[0094] After the video content is generated, video rendering is required in step S7, which mainly involves adding background images and music. Specifically, the real-time rendering and compositing are implemented using a five-layer compositing pipeline, including: a digital human alpha channel video stream layer, a dynamic background template layer, a dynamic subtitle layer, a background music layer, and a global effects layer, and timeline alignment and blending rendering are accelerated by GPU.
[0095] In some alternative implementations, step S7, background replacement includes:
[0096] Based on the category information in the merchant's feature vector, matching background materials are dynamically loaded from the background template library. For example, a hot pot restaurant might use a background of boiling red oil.
[0097] Finally, in step S8, a progressive transmission protocol is adopted to push the video stream to the map interface pop-up for playback within 3 seconds after the user clicks, and the first frame loading time is controlled within 500ms.
[0098] This application automates the entire process from map interaction to video playback, reducing the overall cost of video generation for merchants to 1 / 50 of traditional production, and reducing response latency by 68% compared to cloud solutions, thus solving the real-time problem of personalized digital marketing for offline merchants.
[0099] This application achieves ultimate real-time performance and an immersive user experience, significantly reducing the production cost and barriers to personalized marketing content, and enhancing the dynamic relevance and appeal of marketing content.
[0100] The second aspect of this application provides a real-time generation device for merchant digital human marketing videos based on map interaction, corresponding to the above method, mainly comprising:
[0101] The interactive data acquisition module is used to respond to user interactions on the map interface and capture the selected merchant's POI-ID.
[0102] The merchant feature vector construction module is used to acquire and aggregate multi-source data of the merchant in real time based on the merchant's POI-ID. The multi-source data includes at least basic information, marketing activity data and social media comment data, and constructs a structured merchant feature vector.
[0103] The short video marketing copy generation module is used to dynamically generate short video marketing copy that matches the merchant based on the merchant's feature vector using a large language model.
[0104] The triplet data extraction module is used to simultaneously generate corresponding text, audio stream, and subtitle timeline triplet data based on the marketing copy.
[0105] The digital human matching module is used to automatically match and call the corresponding digital human image from the digital human resource library based on the category information in the merchant feature vector;
[0106] The digital human parameter adjustment module is used to use the voice stream to drive the lip movements of the selected digital human image to synchronize through phoneme mapping technology, and to trigger the preset action template of the digital human based on the keywords in the marketing copy.
[0107] The video rendering module is used to perform real-time rendering and compositing of the driven digital human video stream at the edge computing node. The compositing includes at least: background replacement, overlaying dynamic subtitles corresponding to the subtitle timeline, and adding background music, and finally generating a complete marketing video file.
[0108] The video loading module is used to push the generated marketing video file back to the map interface for playback with low latency via a progressive transmission protocol.
[0109] A third aspect of this application provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method for real-time generation of map-based interactive digital human marketing videos for merchants as described above.
[0110] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A method for real-time generation of digital human marketing videos for merchants based on map interaction, characterized in that, include: S1. Responding to user interaction on the map interface, capture the selected merchant's POI-ID; S2. Based on the merchant's POI-ID, acquire and aggregate the merchant's multi-source data in real time. The multi-source data includes at least basic information, marketing activity data, and social media comment data, and construct a structured merchant feature vector. S3. Based on the merchant feature vector, a large language model dynamically generates short video marketing copy that matches the merchant. S4. Based on the marketing copy, generate corresponding text, audio stream, and subtitle timeline triplet data simultaneously; S5. Based on the category information in the merchant feature vector, automatically match and call the corresponding digital human image from the digital human resource library; S6. Using the voice stream, the selected digital human's lip movements are synchronized through phoneme mapping technology, and the preset action template of the digital human is triggered based on the keywords in the marketing copy. S7. At the edge computing node, the driven digital human video stream is rendered and synthesized in real time. The synthesis includes at least: background replacement, overlaying dynamic subtitles corresponding to the subtitle timeline, and adding background music, and finally generating a complete marketing video file. S8. The generated marketing video file is pushed back to the map interface for playback with low latency via a progressive transmission protocol.
2. The method for real-time generation of merchant digital human marketing videos based on map interaction according to claim 1, characterized in that, In step S2, the real-time acquisition and aggregation of multi-source data is achieved by calling one or more of the following APIs: map service provider API, Dianping API, and merchant's own CRM system API.
3. The method for real-time generation of merchant digital human marketing videos based on map interaction according to claim 1, characterized in that, In step S2, the constructed structured merchant feature vector is represented as follows: V=[a,V_basic,b,V_marketing,c,V_social]; Among them, V_basic is the basic information sub-vector with weight a, V_marketing is the marketing sub-vector with weight b, and V_social is the social sub-vector with weight c. The weights a, b, and c are obtained by looking up a table based on the current time and the merchant category.
4. The method for real-time generation of merchant digital human marketing videos based on map interaction according to claim 1, characterized in that, In step S3, the dynamic generation of short video marketing copy is achieved by the large language model generating copy text with a video length of no more than 15 seconds based on the preset marketing script style template and the merchant feature vector.
5. The method for real-time generation of merchant digital human marketing videos based on map interaction according to claim 1, characterized in that, Step S3 further includes: Step S31: Dynamically construct the prompt word P based on the merchant feature vector V. The construction formula is as follows: P=Template[Keywords(V)]+Constraints; Wherein, Template is a preset marketing script style template; Keywords(V) is a function that extracts core keywords from feature vector V; Constraints are the generation constraints, which include at least video duration constraints and sentence structure constraints; Step S32: Input the prompt word P into the large language model to generate short video marketing copy.
6. The method for real-time generation of merchant digital human marketing videos based on map interaction according to claim 1, characterized in that, In step S6, the keyword-triggered preset action template includes: A keyword-action mapping library is constructed. When a keyword from the mapping library is detected in a marketing copy, the digital human image is automatically triggered to perform the corresponding action, which includes gestures, facial expressions, or body animations.
7. The method for real-time generation of merchant digital human marketing videos based on map interaction according to claim 6, characterized in that, In step S6, when keyword K is identified, the probability P(A|K) of triggering the corresponding action A is calculated by the following formula: P(A|K)=B(K)+η*Score(S); Where B(K) is the preset baseline trigger probability of keyword K in the action mapping library, Score(S) is the overall sentiment score of the current marketing copy S, and η is the sentiment influence factor coefficient.
8. The method for real-time generation of merchant digital human marketing videos based on map interaction according to claim 1, characterized in that, In step S7, background replacement includes: Based on the category information in the merchant feature vector, matching background materials are dynamically loaded from the background template library.
9. A real-time generation device for merchant digital human marketing videos based on map interaction, characterized in that, include: The interactive data acquisition module is used to respond to user interactions on the map interface and capture the selected merchant's POI-ID. The merchant feature vector construction module is used to acquire and aggregate multi-source data of the merchant in real time based on the merchant's POI-ID. The multi-source data includes at least basic information, marketing activity data and social media comment data, and constructs a structured merchant feature vector. The short video marketing copy generation module is used to dynamically generate short video marketing copy that matches the merchant based on the merchant's feature vector using a large language model. The triplet data extraction module is used to simultaneously generate corresponding text, audio stream, and subtitle timeline triplet data based on the marketing copy. The digital human matching module is used to automatically match and call the corresponding digital human image from the digital human resource library based on the category information in the merchant feature vector; The digital human parameter adjustment module is used to use the voice stream to drive the lip movements of the selected digital human image to synchronize through phoneme mapping technology, and to trigger the preset action template of the digital human based on the keywords in the marketing copy. The video rendering module is used to perform real-time rendering and compositing of the driven digital human video stream at the edge computing node. The compositing includes at least: background replacement, overlaying dynamic subtitles corresponding to the subtitle timeline, and adding background music, and finally generating a complete marketing video file. The video loading module is used to push the generated marketing video file back to the map interface for playback with low latency via a progressive transmission protocol.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the real-time generation method for merchant digital human marketing videos based on map interaction as described in any one of claims 1 to 8.