A system for context and attention aware video editing

IN604930BActive Publication Date: 2026-10-05INDIRA GANDHI DELHI TECHNICAL UNIV FOR WOMEN
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
IN202511115712
Authority / Receiving Office
IN · IN
Patent Type
Patents
Current Assignee / Owner
Filing Date
2025-11-22
Publication Date
2026-10-05

AI Technical Summary

Technical Problem

Existing video editing technologies are resource-intensive, lack contextual understanding, and struggle with efficient processing on low-compute devices, leading to suboptimal outputs and inadequate tamper-proof mechanisms.

Method used

A system utilizing intelligent video analysis, attention scoring, and automated effect recommendations, combined with a distributed computing architecture, to provide efficient, tamper-proof, and context-aware video editing on heterogeneous devices.

Benefits of technology

Enables high-quality video editing with reduced resource dependency, supports multiple operation modes, and ensures auditability, while efficiently processing on low-compute devices.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 00000028_0000
    Figure 00000028_0000
Patent Text Reader

Abstract

The present invention relates to a system(100) for context and attention aware video editing. The present invention includes an orchestrator computing device(102), an edge computing device(110), portable display device(108) and a cloud server computer(114). The orchestrator computing device(102) having an orchestrator database(104), an orchestrator processor(106) and a permissioned ledger blockchain database(112). The orchestrator database(104) stores the computer readable instructions and the input video by the user. The orchestrator database(104) having a chunking module and a job manifester module. The chunking module segments the input video into fixed-interval or dynamic chunks using change-point detection on motion / object activity, peak detection on attention score derivatives. The system(100) avoids per-frame large language model or vision-language model inference by using sparse keyframe analysis, deterministic pipelines, and a graph-based experience store, thereby enabling professional video editing on heterogenous low-compute devices. In the distributed variant, hash-locked artifacts and deterministic parameters enable safe parallelization on heterogenous low-compute clusters.
Need to check novelty before this filing date? Find Prior Art

Description

FIELD OF THE INVENTIONThe present invention relates to a system for context and attention aware videoediting. More specifically, the present invention relates to a context and attentionaware video editing system and method that utilizes intelligent analysis of videocontent, attention scoring, emotion alignment, and automated effectrecommendations to produce efficient and professional video outputs.BACKGROUND OF THE INVENTIONVideo content creation and editing have become increasingly prevalent acrossmultiple domains, including social media, professional filmmaking, onlineeducation, and marketing. Traditional video editing processes are often time consuming, labour-intensive, and require significant technical expertise. Manualediting involves frame-by-frame review, careful selection of shots, and preciseapplication of transitions, effects, and audio adjustments.Existing automated video editing tools often rely on predefined templates, genericheuristics, or rule-based systems. Such approaches typically fail to account for thecontextual importance of video segments, viewer attention patterns, or emotionalalignment of content, leading to suboptimal outputs. Furthermore, many currentsystems are resource-intensive, requiring high-performance computinginfrastructure, and struggle with efficiently processing large or high-resolutionvideo files on low-compute or heterogeneous devices.Additionally, current video editing workflows lack robust mechanisms to ensuretamper-proof editing, reproducibility, and auditability of applied changes, which iscritical in collaborative or distributed editing environments. These limitationshighlight the need for improved methods and systems capable of intelligent,efficient, and reliable video editing.US20250285645 discloses Artificial intelligence-based video edit recommendationmethod for semi-automatic video editing by means of computer-humancooperation. Provided is an artificial intelligence-based video edit recommendationmethod for semi-automatic video editing by means of computer-humancooperation. A video edits recommendation method according to an embodiment ofthe present invention divides video content into cut units; infers the situation anddetects an event by analyzing the divided video content, and recommends videoclips required for an editing process in accordance with the situation inferenceresult, event detection result, and genre information. Therefore, by automaticallyrecommending video clips required for video editing on the basis of artificialintelligence, artificial intelligence performs a considerable amount of work requiredfor a step prior to editing work, and an editor only needs to do supplementary workand follow-up work, and thus effort and time required for editing may notably bereduced.WO2025042388A1 discloses Video context aware editing agent. Techniques forusing an artificial intelligent (AI) agent to modify audiovisual data is describedherein. The system can obtain audiovisual data descriptive of one or more segmentsof audiovisual content. Additionally, the system can generate a semantic contextualdescription of each of the one or more segments of audiovisual content, the semanticcontextual description generated based on one or more content understandingmodels. Moreover, the system can provide the semantic contextual description to auser. Furthermore, the system can receive a prompt from the user comprising oneor more user-requested modifications to the semantic contextual description. Thesystem can provide the prompt from the user and the semantic contextualdescription to the one or more content understanding models. Subsequently, thesystem can receive a second semantic contextual description from the one or morecontent understanding models, the second semantic contextual descriptionreflecting the one or more user-requested modifications.US20250246206A1 discloses Ai-enhanced video editing with intermediate datamodel representation and web-based interface. The present invention relates to acomputer-implemented method and system for generating a video editing projectusing artificial intelligence (AI) and machine learning (ML) techniques. Themethod includes processing a collection of video clips to generate text-basedmetadata, receiving selection criteria to identify relevant video clips, and generatinga natural language prompt based on the selection criteria. The prompt, comprisinginstructions and context, is provided to a large language model (LLM), whichprocesses the input and outputs data for constructing a video project data model.The project data model includes timing data for salient snippets within the selectedvideo clips. A dynamic and interactive web-based user interface is rendered tovisually represent the project data model, offering a timeline view and editing toolsfor refining the video project. This system streamlines the video editing process byintegrating AI-driven content analysis with user-directed editing, resulting in atailored video project that aligns with user-defined thematic elements.WO2023106554A1 discloses Artificial intelligence-based video editrecommendation method for semi-automatic video editing by means of computer human cooperation. Provided is an artificial intelligence-based video editrecommendation method for semi-automatic video editing by means of computer human cooperation. A video edit recommendation method according to anembodiment of the present invention divides video content into cut units; infers thesituation and detects an event by analyzing the divided video content, andrecommends video clips required for an editing process in accordance with thesituation inference result, event detection result, and genre information. Therefore,by automatically recommending video clips required for video editing on the basisof artificial intelligence, artificial intelligence performs a considerable amount ofwork required for a step prior to editing work, and an editor only needs to dosupplementary work and follow-up work, and thus effort and time required forediting may notably be reduced.Thus, there is a need for the present invention to address the issues in the existingtechnologies and provide a system for context and attention aware video editing.OBJECTIVE OF THE INVENTIONThe main objective of the present invention is to provide a context and attentionaware video editing system that analyzes video content intelligently to generateoptimized editing suggestions and outputs.Another objective of the present invention is to provide automatedrecommendations for effects, transitions, B-roll, filters, audio enhancements,layouts, music, and typography to improve video quality and viewer engagement.Yet another objective of the present invention is to provide multiple modes ofoperation including manual, semi-automated, and fully automated editing underuser-defined constraints for flexibility and control.Yet another objective of the present invention is to provide a tamper-proof andauditable system that records editing operations, job manifests, and metadata in asecure and immutable manner.Yet another objective of the present invention is to provide efficient processing onheterogeneous or low-compute devices by leveraging keyframe analysis, attentionscoring, semantic summarization, and graph-based experience stores.Yet another objective of the present invention is to reduce dependency on resource intensive per-frame analysis while maintaining high-quality professional videooutputs.SUMMARY OF THE INVENTIONThe present invention relates to a system for context and attention aware videoediting. The present invention includes an orchestrator computing device, an edgecomputing device, a portable display device and a cloud server computer. Theorchestrator computing device having an orchestrator database, an orchestratorprocessor and a permissioned ledger blockchain database. The orchestratordatabase stores the computer readable instructions and the input video by the user.The orchestrator database having a chunking module and a job manifester module.The chunking module segments the input video into fixed-interval or dynamicchunks using change-point detection on motion / object activity, or peak detection onattention score derivatives. The job manifester module is used to create jobmanifests file for each chunk of the input video. The orchestrator processor executescomputer-readable instructions stored in the orchestrator database to execute thechunking module and the job manifester module. The permissioned ledgerblockchain database stores the job / asset hashes and deterministic parameters arerecorded using the blockchain technology to enable the system to detect tamperingand replays. The edge computing device is connected to the orchestrator computingdevice. The edge computing device having an edge computing device database andan edge computing device processor. The edge computing device database storesthe computer readable instructions. The edge computing device database having afeature extraction module, an emotion prior module, an attention scoring module, asemantic summarization module, a decision module, a modifications file, a postprocessing engine and a mode controller module. The feature extraction modulecomputes per-chunk component scores including scores of saliency, motion, andobject presence and each score undergoes min-max normalization across the videosequence. The emotion prior module queries a lightweight language model toproduce an emotion alignment score and non-negative weights. The attentionscoring module calculates the attention scores of each chunk by fusing thenormalized scores of the feature extraction module with the output of the emotionprior module and flags chunks with low attention scores for augmentation. Thesemantic summarization module calculates textual synopsis by summarizing threekeyframes (begin / mid / end) of each chunk using sparse vision-language models.The decision module integrates the attention score, the textual synopsis, and agraph-based experience store to recommend targeted insertions including hooks, Broll, filters, transitions, and sounds, as well as global effects including layouttemplates, music, and typography. The modifications file stores the decisions thatare taken by the decision module in a structured JSON schema specifying chunkidentifiers, effect type, reasons (string), and type specific fields. The postprocessing engine is a deterministic engine which applies filters to each chunk andimplements inter-chunk transitions and deterministic modifications usingFFmpeg / OpenCV with audio-aware smoothing, cross-fades, and loudnessnormalization. The mode controller module supports manual, semi-automated andautomated execution of the modifications file subject to user policy constraints. Theedge computing device processor is used to execute computer readable instructionsand execute the jobs assigned to the edge computing device. The orchestratorcomputing device sends the job manifest file for each chunk of the input video tothe edge computing device. The portable display device is connected to theorchestrator computing device and edge computing device. The portable displaydevice is used to preview modifications, display effect recommendations, andrender final video outputs. The cloud server computer is connected to theorchestrator computing device and edge computing device. The cloud servercomputer is used for optional audio enhancement and / or ASR to assist qualitycontrol without requiring per-frame video inference. The edge computing deviceadvertise codecs, CPU features (e.g., AVX2 / NEON), memory and bandwidth andthe job manifester module of the orchestrator computing device co-locates jobs withcached fragments when possible, which helps the system in assigning jobs to theedge computing device which are available in a distributed computing environmentand instead of blindly sending tasks, the system matches the needs of the task withthe capabilities of the edge computing device and take into consideration where therelevant data might already be stored. The system avoids per-frame large languagemodel or vision-language model inference thereby enabling professional videoediting on heterogenous low-compute devices.The main advantage of the present invention is to provide a context and attentionaware video editing system that analyzes video content intelligently to generateoptimized editing suggestions and outputs.Another advantage of the present invention is to provide automatedrecommendations for effects, transitions, B-roll, filters, audio enhancements,layouts, music, and typography to improve video quality and viewer engagement.Yet another advantage of the present invention is to provide multiple modes ofoperation including manual, semi-automated, and fully automated editing underuser-defined constraints for flexibility and control.Yet another advantage of the present invention is to provide a tamper-proof andauditable system that records editing operations, job manifests, and metadata in asecure and immutable manner.Yet another advantage of the present invention is to provide efficient processing onheterogeneous or low-compute devices by leveraging keyframe analysis, attentionscoring, semantic summarization, and graph-based experience stores.Yet another advantage of the present invention is to reduce dependency on resource intensive per-frame analysis while maintaining high-quality professional videooutputs.BRIEF DESCRIPTION OF THE INVENTIONThe accompanying drawings are incorporated in and constitute a part of thisspecification to provide a further understanding of the invention. The drawingsillustrate one embodiment of the invention and together with the description, serveto explain the principles of the invention.Fig.1: Illustrates block diagram of a system for context and attention aware videoediting.Fig.2: Illustrates distributed pipeline with orchestrator and ledger.Fig.3: Illustrates Mode control with distributed execution.Fig.4: Illustrates Distributed chunk job sequence with ledger receipts.Fig.5: Illustrates Scheduler internals: locality-aware assignment and stragglerhandling.DETAILED DESCRIPTION OF THE INVENTIONDefinitionThe terms "a" or "an", as used herein, are defined as one or as more than one. Theterm "plurality", as used herein, is defined as two as or more than two. The term"another", as used herein, is defined as at least a second or more. The terms"including" and / or "having", as used herein, are defined as comprising (i.e., openlanguage). The term "coupled", as used herein, is defined as connected, althoughnot necessarily directly, and not necessarily mechanically.The term "comprising" is not intended to limit inventions to only claiming thepresent invention with such comprising language. Any invention using the termcomprising could be separated into one or more claims using "consisting" or"consisting of" claim language and is so intended. The term "comprising" is usedinterchangeably used by the terms "having" or "containing".Reference throughout this document to "one embodiment", "certain embodiments","an embodiment", "another embodiment", and "yet another embodiment" or similarterms means that a particular feature, structure, or characteristic described inconnection with the embodiment is included in at least one embodiment of thepresent invention. Thus, the appearances of such phrases or in various placesthroughout this specification are not necessarily all referring to the sameembodiment. Furthermore, the particular features, structures, or characteristics arecombined in any suitable manner in one or more embodiments without limitation.The term "or" as used herein is to be interpreted as an inclusive or meaning any oneor any combination. Therefore, "A, B or C" means any of the following: "A; B; C;A and B; A and C; B and C; A, B and C". An exception to this definition will occuronly when a combination of elements, functions, steps, or acts are in some wayinherently mutually exclusive.As used herein, the term "one or more" generally refers to, but not limited to,singular as well as the plural form of the term.The drawings featured in the figures are to illustrate certain convenientembodiments of the present invention and are not to be considered as a limitationto that. Term "means" preceding a present participle of operation indicates thedesired function for which there is one or more embodiments, i.e., one or moremethods, devices, or apparatuses for achieving the desired function and that oneskilled in the art could select from these or their equivalent in view of the disclosureherein and use of the term "means" is not intended to be limiting disclosure hereinand use of the term "means" is not intended to be limiting.Fig.1: Illustrates block diagram of a system for context and attention aware videoediting. The present invention includes an orchestrator computing device (102),edge computing device (110), a portable display device (108) and a cloud servercomputer (114). The orchestrator computing device (102) includes an orchestratordatabase (104), an orchestrator processor (106) and a permissioned ledgerblockchain database (112). The edge computing device (110) includes an edgecomputing device database (116) and an edge computing device processor (118).The portable display device (108) is connected to the orchestrator computing device(102) and edge computing device (110). The cloud server computer (114) isconnected to the orchestrator computing device (102).Fig.2: Illustrates distributed pipeline with orchestrator and ledger. The input videois segmented into fixed-interval or dynamic chunks by executing the chunkingmodule of the orchestrator computing device (102). The job manifest file isgenerated for each chunk of the input video using a job manifester module of theorchestrator computing device (102). The job manifest file is transmitted to the edgecomputing device (110) which is connected to the orchestrator computing device(102). Features are extracted from each chunk by computing per-chunk saliency,motion, and object presence scores, and applying min-max normalization to eachscore across the video sequence. An emotion alignment score is derived and nonnegative weights by querying an emotion prior module using a lightweight languagemodel. Attention scores are calculated for each chunk by fusing the normalizedfeature scores with the emotion alignment score, and flagging low-attention chunksfor augmentation. Each chunk is summarized by producing a textual synopsis fromthree representative keyframes (beginning, middle, and end) using a semanticsummarization module. The attention scores, textual synopses, and a graph-basedexperience store are integrated within a decision module to recommend targetedeffects including hooks, B-roll, filters, transitions, and sounds, as well as globaleffects including layout templates, music, and typography. Recording saidrecommendations in a structured modifications file in JSON schema formatspecifying chunk identifiers, effect types, reasons, and type-specific fields. The saidmodifications are applied deterministically with a post-processing engine usingFFmpeg or OpenCV, the post-processing engine implementing inter-chunktransitions, audio-aware smoothing, cross-fades, and loudness normalization.Effects are applied with audio-aware stitching. The modifications are executedunder multiple modes including manual mode, semi-automated mode with perchunk confirmation or bulk rules, and automated mode under user policyconstraints using the mode controller module of the edge computing device (110).Job manifest hashes, worker identities, artifact hashes, and deterministic parametersare stored on a permissioned ledger blockchain database (112) to provideauditability, immutability, and tamper detection. Previews of modifications, effectrecommendations, and final video outputs are displayed on a portable displaydevice (108). Audio quality is optionally enhanced and / or automatic speechrecognition (ASR) is performed via a cloud server computer (114) without requiringper-frame video inference.Fig.3: Illustrates Mode control with distributed execution. The modifications areexecuted under multiple modes including manual mode, semi-automated mode withper-chunk confirmation or bulk rules, and automated mode under user policyconstraints using the mode controller module of the edge computing device (110).Fig.4: Illustrates Distributed chunk job sequence with ledger receipts. The inputvideo is segmented into fixed-interval or dynamic chunks by executing thechunking module of the orchestrator computing device (102). The job manifest fileis generated for each chunk of the input video using a job manifester module of theorchestrator computing device (102). The job manifest file is transmitted to the edgecomputing device (110) which is connected to the orchestrator computing device(102). Features are extracted from each chunk by computing per-chunk saliency,motion, and object presence scores, and applying min-max normalization to eachscore across the video sequence. An emotion alignment score is derived and nonnegative weights by querying an emotion prior module using a lightweight languagemodel. Attention scores are calculated for each chunk by fusing the normalizedfeature scores with the emotion alignment score, and flagging low-attention chunksfor augmentation. Each chunk is summarized by producing a textual synopsis fromthree representative keyframes (beginning, middle, and end) using a semanticsummarization module. The attention scores, textual synopses, and a graph-basedexperience store are integrated within a decision module to recommend targetedeffects including hooks, B-roll, filters, transitions, and sounds, as well as globaleffects including layout templates, music, and typography. Recording saidrecommendations in a structured modifications file in JSON schema formatspecifying chunk identifiers, effect types, reasons, and type-specific fields. The saidmodifications are applied deterministically with a post-processing engine usingFFmpeg or OpenCV, the post-processing engine implementing inter-chunktransitions, audio-aware smoothing, cross-fades, and loudness normalization.Effects are applied with audio-aware stitching. The modifications are executedunder multiple modes including manual mode, semi-automated mode with perchunk confirmation or bulk rules, and automated mode under user policyconstraints using the mode controller module of the edge computing device (110).Job manifest hashes, worker identities, artifact hashes, and deterministic parametersare stored on a permissioned ledger blockchain database (112) to provideauditability, immutability, and tamper detection. Previews of modifications, effectrecommendations, and final video outputs are displayed on a portable displaydevice (108). Audio quality is optionally enhanced and / or automatic speechrecognition (ASR) is performed via a cloud server computer (114) without requiringper-frame video inference.Fig.5: Illustrates Scheduler internals: locality-aware assignment and stragglerhandling. The input video is segmented into fixed-interval or dynamic chunks byexecuting the chunking module of the orchestrator computing device (102). The jobmanifest file is generated for each chunk of the input video using a job manifestermodule of the orchestrator computing device (102). The job manifest file istransmitted to the edge computing device (110) which is connected to theorchestrator computing device (102). Features are extracted from each chunk bycomputing per-chunk saliency, motion, and object presence scores, and applyingmin-max normalization to each score across the video sequence. An emotionalignment score is derived and non-negative weights by querying an emotion priormodule using a lightweight language model. Attention scores are calculated foreach chunk by fusing the normalized feature scores with the emotion alignmentscore, and flagging low-attention chunks for augmentation. Each chunk issummarized by producing a textual synopsis from three representative keyframes(beginning, middle, and end) using a semantic summarization module. Theattention scores, textual synopses, and a graph-based experience store are integratedwithin a decision module to recommend targeted effects including hooks, B-roll,filters, transitions, and sounds, as well as global effects including layout templates,music, and typography. Recording said recommendations in a structuredmodifications file in JSON schema format specifying chunk identifiers, effect types,reasons, and type-specific fields. The said modifications are applieddeterministically with a post-processing engine using FFmpeg or OpenCV, the postprocessing engine implementing inter-chunk transitions, audio-aware smoothing,cross-fades, and loudness normalization. Effects are applied with audio-awarestitching. The modifications are executed under multiple modes including manualmode, semi-automated mode with per-chunk confirmation or bulk rules, andautomated mode under user policy constraints using the mode controller module ofthe edge computing device (110). Job manifest hashes, worker identities, artifacthashes, and deterministic parameters are stored on a permissioned ledgerblockchain database (112) to provide auditability, immutability, and tamperdetection. Previews of modifications, effect recommendations, and final videooutputs are displayed on a portable display device (108). Audio quality is optionallyenhanced and / or automatic speech recognition (ASR) is performed via a cloudserver computer (114) without requiring per-frame video inference. Theorchestrator computing device (102) verifies results and records the receipts in thepermissioned ledger blockchain database (112) for auditability. The orchestratorcomputing device (102) triggers speculative execution of a chunk-level job on analternative edge computing device (110) if the previous edge computing device(110) fails to return a timely receipt.The present invention relates to a system for context and attention aware videoediting. The present invention includes an orchestrator computing device, an edgecomputing device, a portable display device and a cloud server computer. Theorchestrator computing device having an orchestrator database, an orchestratorprocessor and a permissioned ledger blockchain database. In an embodiment, theorchestrator computing device, shards the input video into chunk-level jobmanifests containing source fragment references, modification subsets,deterministic parameters, and expected artifact names and hashes, QoS targets (e.g.,max loudness error, transition latency) and location hints for data locality and cachereuse. The orchestrator database stores the computer readable instructions and theinput video by the user. The orchestrator database having a chunking module and ajob manifester module. The chunking module segments the input video into fixed interval or dynamic chunks using change-point detection on motion / object activity,or peak detection on attention score derivatives. The job manifester module is usedto create job manifests file for each chunk of the input video. The orchestratorprocessor executes computer-readable instructions stored in the orchestratordatabase to execute the chunking module and the job manifester module. Thepermissioned ledger blockchain database stores the job / asset hashes anddeterministic parameters are recorded using the blockchain technology to enablethe system to detect tampering and replays. The edge computing device isconnected to the orchestrator computing device. The edge computing device havingan edge computing device database and an edge computing device processor. Theedge computing device database stores the computer readable instructions. Theedge computing device database having a feature extraction module, an emotionprior module, an attention scoring module, a semantic summarization module, adecision module, a modifications file, a post-processing engine and a modecontroller module. The feature extraction module computes per-chunk componentscores including scores of saliency, motion, and object presence and each scoreundergoes min-max normalization across the video sequence. The emotion priormodule queries a lightweight language model to produce an emotion alignmentscore and non-negative weights. The attention scoring module calculates theattention scores of each chunk by fusing the normalized scores of the featureextraction module with the output of the emotion prior module and flags chunkswith low attention scores for augmentation. The semantic summarization modulecalculates textual synopsis by summarizing three keyframes (begin / mid / end) ofeach chunk using sparse vision-language models. The decision module integratesthe attention score, the textual synopsis, and a graph-based experience store torecommend targeted insertions including hooks, B-roll, filters, transitions, andsounds, as well as global effects including layout templates, music, and typography.The modifications file stores the decisions that are taken by the decision module ina structured JSON schema specifying chunk identifiers, effect type, reasons (string),and type specific fields. The post-processing engine is a deterministic engine whichapplies filters to each chunk and implements inter-chunk transitions anddeterministic modifications using FFmpeg / OpenCV with audio-aware smoothing,cross-fades, and loudness normalization. The mode controller module supportsmanual, semi-automated and automated execution of the modifications file subjectto user policy constraints. The edge computing device processor is used to executecomputer readable instructions and execute the jobs assigned to the edge computingdevice. The orchestrator computing device sends the job manifest file for eachchunk of the input video to the edge computing device. In an embodiment, jobreceipts comprising job manifest hash, artifact hashes, quality-of-service metricsand worker identity are immutably appended to the permissioned ledger blockchaindatabase using BFT or PoA style consensus without requiring any cryptocurrency.The portable display device is connected to the orchestrator computing device andedge computing device. The portable display device is used to previewmodifications, display effect recommendations, and render final video outputs. Thecloud server computer is connected to the orchestrator computing device and edgecomputing device. The cloud server computer is used for optional audioenhancement and / or ASR to assist quality control without requiring per-frame videoinference. The edge computing device advertise codecs, CPU features (e.g.,AVX2 / NEON), memory and bandwidth and the job manifester module of theorchestrator computing device co-locates jobs with cached fragments whenpossible, which helps the system in assigning jobs to the edge computing devicewhich are available in a distributed computing environment and instead of blindlysending tasks, the system matches the needs of the task with the capabilities of theedge computing device and take into consideration where the relevant data mightalready be stored. The system avoids per-frame large language model or vision language model inference by using sparse keyframe analysis, deterministicpipelines, and a graph-based experience store, thereby enabling professional videoediting on heterogenous low-compute devices. In the distributed variant, hash locked artifacts and deterministic parameters enable safe parallelization onheterogenous low-compute clusters.In an embodiment, the present invention relates to a system for context and attentionaware video editing. The present invention includes a one or more orchestratorcomputing devices, a one or more edge computing devices, a portable displaydevice and a cloud server computer. The one or more orchestrator computingdevices having an orchestrator database, an orchestrator processor and apermissioned ledger blockchain database. In an embodiment, the one or moreorchestrator computing devices, shards the input video into chunk-level jobmanifests containing source fragment references, modification subsets,deterministic parameters, and expected artifact names and hashes, QoS targets (e.g.,max loudness error, transition latency) and location hints for data locality and cachereuse. The orchestrator database stores the computer readable instructions and theinput video by the user. The orchestrator database having a chunking module and ajob manifester module. The chunking module segments the input video into fixedinterval or dynamic chunks using change-point detection on motion / object activity,and peak detection on attention score derivatives. The job manifester module is usedto create job manifests file for each chunk of the input video. The orchestratorprocessor executes computer-readable instructions stored in the orchestratordatabase to execute the chunking module and the job manifester module. Thepermissioned ledger blockchain database stores the job / asset hashes anddeterministic parameters are recorded using the blockchain technology to enablethe system to detect tampering and replays. The one or more edge computingdevices are connected to the one or more orchestrator computing devices. The oneor more edge computing devices having an edge computing device database and anedge computing device processor. The edge computing device database stores thecomputer readable instructions. The edge computing device database having afeature extraction module, an emotion prior module, an attention scoring module, asemantic summarization module, a decision module, a modifications file, a postprocessing engine and a mode controller module. The feature extraction modulecomputes per-chunk component scores including scores of saliency, motion, andobject presence and each score undergoes min-max normalization across the videosequence. The emotion prior module queries a lightweight language model toproduce an emotion alignment score and non-negative weights. The attentionscoring module calculates the attention scores of each chunk by fusing thenormalized scores of the feature extraction module with the output of the emotionprior module and flags chunks with low attention scores for augmentation. Thesemantic summarization module calculates textual synopsis by summarizing threekeyframes (begin / mid / end) of each chunk using sparse vision-language models.The decision module integrates the attention score, the textual synopsis, and agraph-based experience store to recommend targeted insertions including hooks, Broll, filters, transitions, and sounds, as well as global effects including layouttemplates, music, and typography. The modifications file stores the decisions thatare taken by the decision module in a structured JSON schema specifying chunkidentifiers, effect type, reasons (string), and type specific fields. The postprocessing engine is a deterministic engine which applies filters to each chunk andimplements inter-chunk transitions and deterministic modifications usingFFmpeg / OpenCV with audio-aware smoothing, cross-fades, and loudnessnormalization. The mode controller module supports manual, semi-automated andautomated execution of the modifications file subject to user policy constraints. Theedge computing device processor is used to execute computer readable instructionsand execute the jobs assigned to the one or more edge computing devices. The oneor more orchestrator computing devices send the job manifest file for each chunkof the input video to the one or more edge computing devices. In an embodiment,job receipts comprising job manifest hash, artifact hashes, quality-of-servicemetrics and worker identity are immutably appended to the permissioned ledgerblockchain database using BFT or PoA style consensus without requiring anycryptocurrency. The portable display device is connected to the one or moreorchestrator computing devices and the one or more edge computing devices. Theportable display device is used to preview modifications, display effectrecommendations, and render final video outputs. The cloud server computer isconnected to the one or more orchestrator computing devices and the one or moreedge computing devices. The cloud server computer is used for optional audioenhancement and / or ASR to assist quality control without requiring per-frame videoinference. The one or more edge computing devices advertise codecs, CPU features(e.g., AVX2 / NEON), memory and bandwidth and the job manifester module of theone or more orchestrator computing devices co-locates jobs with cached fragmentswhen possible, which helps the system in assigning jobs to the one or more edgecomputing devices which are available in a distributed computing environment andinstead of blindly sending tasks, the system matches the needs of the task with thecapabilities of the one or more edge computing devices and take into considerationwhere the relevant data might already be stored. The system avoids per-frame largelanguage model or vision-language model inference by using sparse keyframeanalysis, deterministic pipelines, and a graph-based experience store, therebyenabling professional video editing on heterogenous low-compute devices. In thedistributed variant, hash-locked artifacts and deterministic parameters enable safeparallelization on heterogenous low-compute clusters.In an embodiment the present invention relates to a method for context and attentionaware video editing by using the system, the method includes:segmenting an input video into fixed-interval or dynamic chunks by executing thechunking module of the orchestrator computing device;generating a job manifest file for each chunk of the input video using a jobmanifester module of the orchestrator computing device;transmitting the job manifest file to edge computing device connected to theorchestrator computing device;extracting features from each chunk by computing per-chunk saliency, motion, andobject presence scores, and applying min-max normalization to each score acrossthe video sequence;deriving an emotion alignment score and non-negative weights by querying anemotion prior module using a lightweight language model;calculating attention scores for each chunk by fusing the normalized feature scoreswith the emotion alignment score, and flagging low-attention chunks foraugmentation;summarizing each chunk by producing a textual synopsis from three representativekeyframes (beginning, middle, and end) using a semantic summarization module;integrating the attention scores, textual synopses, and a graph-based experiencestore within a decision module to recommend targeted effects including hooks, Broll, filters, transitions, and sounds, as well as global effects including layouttemplates, music, and typography;recording said recommendations in a structured modifications file in JSON schemaformat specifying chunk identifiers, effect types, reasons, and type-specific fields;applying said modifications deterministically with a post-processing engine usingFFmpeg or OpenCV, the post-processing engine implementing inter-chunktransitions, audio-aware smoothing, cross-fades, and loudness normalization;applying effects with audio-aware stitching;executing the modifications under multiple modes including manual mode, semiautomated mode with per-chunk confirmation or bulk rules, and automated modeunder user policy constraints using the mode controller module of the edgecomputing device;storing job manifest hashes, worker identities, artifact hashes, and deterministicparameters on a permissioned ledger blockchain database to provide auditability,immutability, and tamper detection;displaying previews of modifications, effect recommendations, and final videooutputs on a portable display device; andoptionally enhancing audio quality and / or performing automatic speech recognition(ASR) via a cloud server computer without requiring per-frame video inference.In an embodiment, the B-roll insertion preserves speech continuity via timewarping and cross-fading.In an embodiment, the orchestrator computing device verifies results and recordsthe receipts in the permissioned ledger blockchain database for auditability.In an embodiment, the orchestrator computing device triggers speculative executionof a chunk-level job on an alternative edge computing device if the previous edgecomputing device fails to return a timely receipt.In an embodiment, effects, content descriptors, and editing outcomes are stored ina heterogeneous graph database, and wherein the method further comprisesperforming community detection on said heterogeneous graph using a Louvainclustering algorithm to generate clusters of editing patterns, the clusters beingemployed as priors to regularize and bias future effect recommendations towardcontextually successful outcomes.In an embodiment the present invention relates to a method for context and attentionaware video editing by using the system, the method includes:segmenting an input video into fixed-interval or dynamic chunks by executing thechunking module of the orchestrator computing device;generating a job manifest file for each chunk of the input video using a jobmanifester module of the orchestrator computing device;transmitting the job manifest file to the one or more computing devices connectedto the orchestrator computing device;extracting features from each chunk by computing per-chunk saliency, motion, andobject presence scores, and applying min-max normalization to each score acrossthe video sequence;deriving an emotion alignment score and non-negative weights by querying anemotion prior module using a lightweight language model;calculating attention scores for each chunk by fusing the normalized feature scoreswith the emotion alignment score, and flagging low-attention chunks foraugmentation;summarizing each chunk by producing a textual synopsis from three representativekeyframes (beginning, middle, and end) using a semantic summarization module;integrating the attention scores, textual synopses, and a graph-based experiencestore within a decision module to recommend targeted effects including hooks, Broll, filters, transitions, and sounds, as well as global effects including layouttemplates, music, and typography;recording said recommendations in a structured modifications file in JSON schemaformat specifying chunk identifiers, effect types, reasons, and type-specific fields;applying said modifications deterministically with a post-processing engine usingFFmpeg or OpenCV, the post-processing engine implementing inter-chunktransitions, audio-aware smoothing, cross-fades, and loudness normalization;applying effects with audio-aware stitching;executing the modifications under multiple modes including manual mode, semiautomated mode with per-chunk confirmation or bulk rules, and automated modeunder user policy constraints using the mode controller module of the one or moreedge computing devices;storing job manifest hashes, worker identities, artifact hashes, and deterministicparameters on a permissioned ledger blockchain database to provide auditability,immutability, and tamper detection;displaying previews of modifications, effect recommendations, and final videooutputs on a portable display device; andoptionally enhancing audio quality and / or performing automatic speech recognition(ASR) via a cloud server computer (114) without requiring per-frame videoinference.In an embodiment, the B-roll insertion preserves speech continuity via time warping and cross-fading.In an embodiment, the one or more orchestrator computing devices verifies resultsand records the receipts in the permissioned ledger blockchain database forauditability.In an embodiment, the orchestrator computing device triggers speculative executionof a chunk-level job on an alternative the one or more edge computing devices ifthe previous the one or more edge computing devices fails to return a timely receipt.In an embodiment, effects, content descriptors, and editing outcomes are stored ina heterogeneous graph database, and wherein the method further comprisesperforming community detection on said heterogeneous graph using a Louvainclustering algorithm to generate clusters of editing patterns, the clusters beingemployed as priors to regularize and bias future effect recommendations towardcontextually successful outcomes.

Claims

1. A system (100) for context and attention aware video editing, the system (100) comprising: an at least one orchestrator computing device (102), the at least one orchestrator computing device (102) having an orchestrator database (104), the orchestrator database (104) stores the computer readable instructions and the input video by the user, the orchestrator database (104) having a chunking module, the chunking module segments the input video into fixed-interval or dynamic chunks using change-point detection on motion / object activity, or peak detection on attention score derivatives, and a job manifester module, the job manifester module is used to create job manifests file for each chunk of the input video, an orchestrator processor (106), the orchestrator processor (106) executes computer-readable instructions stored in the orchestrator database (104) to execute the chunking module and the job manifester module, and a permissioned ledger blockchain database (112), the permissioned ledger blockchain database (112) stores the job / asset hashes and deterministic parameters are recorded using the blockchain technology to enable the system (100) to detect tampering and replays; an at least one edge computing device (110), the at least one edge computing device (110) is connected to the at least one orchestrator computing device (102), the at least one edge computing device (110) having an edge computing device database (116), the edge computing device database (116) stores the computer readable instructions, the edge computing device database (116) having a feature extraction module, the feature extraction module computes per-chunk component scores including scores of saliency, motion, and object presence and each score undergoes min-max normalization across the video sequence, an emotion prior module, the emotion prior module queries a lightweight language model to produce an emotion alignment score and non-negative weights, an attention scoring module, the attention scoring module calculates the attention scores of each chunk by fusing the normalized scores of the feature extraction module with the output of the emotion prior module and flags chunks with low attention scores for augmentation, a semantic summarization module, the semantic summarization module calculates textual synopsis by summarizing three keyframes (begin / mid / end) of each chunk using sparse vision-language models, a decision module, the decision module integrates the attention score, the textual synopsis, and a graph-based experience store to recommend targeted insertions comprising hooks, B-roll, filters, transitions, and sounds, as well as global effects including layout templates, music, and typography, a modifications file, the modifications file stores the decisions that are taken by the decision module in a structured JSON schema specifying chunk identifiers, effect type, reasons (string), and type specific fields, a post-processing engine, the post-processing engine is a deterministic engine which applies filters to each chunk and implements inter-chunk transitions and deterministic modifications using FFmpeg / OpenCV with audio-aware smoothing, cross-fades, and loudness normalization, a mode controller module, the mode controller module supports manual, semi-automated and automated execution of the modifications file subject to user policy constraints, an edge computing device processor (118), the edge computing device processor (118) is used to execute computer readable instructions and execute the jobs assigned to the at least one edge computing device (110); wherein, the at least one orchestrator computing device (102) sends the job manifest file for each chunk of the input video to the at least one edge computing device (110), a portable display device (108), the portable display device (108) is connected to the at least one orchestrator computing device (102) and at least one edge computing device (110), the portable display device (108) is used to preview modifications, display effect recommendations, and render final video outputs; a cloud server computer (114), the cloud server computer (114) is connected to the at least one orchestrator computing device (102) and at least one edge computing device (110); wherein, the cloud server computer (114) is used for optional audio enhancement and / or ASR to assist quality control without requiring per-frame video inference, wherein, the at least one edge computing device (110) advertise codecs, CPU features (e.g., AVX2 / NEON), memory and bandwidth and the job manifester module of the at least one orchestrator computing device (102) co-locates jobs with cached fragments when possible, which helps the system (100) in assigning jobs to the at least one edge computing device (110) which are available in a distributed computing environment and instead of blindly sending tasks, the system (100) matches the needs of the task with the capabilities of the at least one edge computing device (110) and take into consideration where the relevant data might already be stored, characterize in that, the system (100) avoids per-frame large language model or vision-language model inference by using sparse keyframe analysis, deterministic pipelines, and a graph-based experience store, thereby enabling professional video editing on heterogenous low-compute devices, characterize in that, in the distributed variant, hash-locked artifacts and deterministic parameters enable safe parallelization on heterogenous low-compute clusters.

2. The system (100) as claimed in claim 1, wherein the at least one orchestrator computing device (102), shards the input video into chunk-level job manifests containing source fragment references, modification subsets, deterministic parameters, and expected artifact names and hashes, QoS targets (e.g., max loudness error, transition latency) and location hints for data locality and cache reuse.

3. The system (100) as claimed in claim 1, wherein job receipts comprising job manifest hash, artifact hashes, quality-of-service metrics and worker identity are immutably appended to the permissioned ledger blockchain database (112) using BFT or PoA style consensus without requiring any cryptocurrency.

4. A method for context and attention aware video editing by using the system (100) as claimed in claim 1, the method comprising: segmenting an input video into fixed-interval or dynamic chunks by executing the chunking module of the orchestrator computing device (102); generating a job manifest file for each chunk of the input video using a job manifester module of the orchestrator computing device (102); transmitting the job manifest file to at least one edge computing device (110) connected to the orchestrator computing device (102); extracting features from each chunk by computing per-chunk saliency, motion, and object presence scores, and applying min-max normalization to each score across the video sequence; deriving an emotion alignment score and non-negative weights by querying an emotion prior module using a lightweight language model; calculating attention scores for each chunk by fusing the normalized feature scores with the emotion alignment score, and flagging low-attention chunks for augmentation; summarizing each chunk by producing a textual synopsis from three representative keyframes (beginning, middle, and end) using a semantic summarization module; integrating the attention scores, textual synopses, and a graph-based experience store within a decision module to recommend targeted effects including hooks, Broll, filters, transitions, and sounds, as well as global effects including layout templates, music, and typography; recording said recommendations in a structured modifications file in JSON schema format specifying chunk identifiers, effect types, reasons, and type-specific fields; applying said modifications deterministically with a post-processing engine using FFmpeg or OpenCV, the post-processing engine implementing inter-chunk transitions, audio-aware smoothing, cross-fades, and loudness normalization; applying effects with audio-aware stitching; executing the modifications under multiple modes including manual mode, semiautomated mode with per-chunk confirmation or bulk rules, and automated mode under user policy constraints using the mode controller module of the at least one edge computing device (110); storing job manifest hashes, worker identities, artifact hashes, and deterministic parameters on a permissioned ledger blockchain database (112) to provide auditability, immutability, and tamper detection; displaying previews of modifications, effect recommendations, and final video outputs on a portable display device (108); and optionally enhancing audio quality and / or performing automatic speech recognition (ASR) via a cloud server computer (114) without requiring per-frame video inference.

5. The method as claimed in claim 4, wherein B-roll insertion preserves speech continuity via time-warping and cross-fading.

6. The method as claimed in claim 4, wherein the at least one orchestrator computing device (102) verifies results and records the receipts in the permissioned ledger blockchain database (112) for auditability.

7. The method as claimed in claim 4, wherein the orchestrator computing device (102) triggers speculative execution of a chunk-level job on an alternative at least one edge computing device (110) if the previous at least one edge computing device (110) fails to return a timely receipt.

8. The method as claimed in claim 4, wherein effects, content descriptors, and editing outcomes are stored in a heterogeneous graph database, and wherein the method further comprises performing community detection on said heterogeneous graph using a Louvain clustering algorithm to generate clusters of editing patterns, the clusters being employed as priors to regularize and bias future effect recommendations toward contextually successful outcomes.