Text-to-video and image-to-video
Generate video clips from text prompts or reference images for creative content, ad assets and previsualization.
AI Info Center
Berserk is Glacier AI's core in-house model series. AI Tools provides task tools, and the Aggregation Hub provides model access.
The in-house Berserk model series
Includes chat, text-to-image, text-to-video and other single-modal and cross-modal models.
- Berserk-V4flash: benchmarked against DeepSeek V4 flash.
- Berserk-5.2: benchmarked against GLM5.2.
- Berserk-H3: benchmarked against MiniMax H3.
Positioning: results close to the benchmark model, with a token price of 50% of the corresponding benchmark's lowest market price of the day. Reference prices are updated or may change daily; the actual price is the day's quote. This describes the pricing rule, not a live market quote.
Aggregation Hub
Unified access to Chinese models, with one protocol and keys and calls enabled per project.
Integrations cover Qwen, ERNIE, Zhipu, Kimi, DeepSeek, Doubao, Hunyuan and Spark; specific model versions and APIs are confirmed at integration.
AI Tools
84 features in six categories, covering video, text, images, speech, cross-modal, time series and perception. Super-resolution, enhancement, AI denoising and bitrate recovery are part of AI Tools.
View original image and video demos ↗
Full feature catalog
Video super-resolution and frame interpolation
Upscale video resolution and frame rate for smoother, sharper footage, ideal for restoring old films and slow motion.
Video highlight detection and auto-editing
Automatically find the best moments and cut them into short clips for sports, live streams and short-form video.
Video subtitle removal
Detect and remove burned-in subtitles for re-dubbing and multilingual distribution.
Video watermark removal
Locate and remove watermarks and channel logos for asset cleanup and compliance.
Video stabilization
Remove handheld camera shake for steady footage and a better viewing experience.
Video matting and background replacement
Separate people in the foreground from the background, enabling virtual backgrounds without a green screen.
Action and event recognition
Recognize human actions and events in video for security, sports and industrial monitoring.
Activity, fall and fatigue detection
Detect falls, fatigue and abnormal activity in elder care, construction sites and driving, and raise alerts.
Visual anomaly detection
Spot deviations from normal patterns in video for production lines, inspections and security.
Multi-object tracking
Continuously track multiple objects in video and output track IDs for pedestrian and traffic flow analysis.
Object detection and multi-object tracking
Detect objects in video and images and track them over time, powering counting, speed measurement and behavior analysis.
Trajectory prediction and motion planning
Predict the future paths of pedestrians, vehicles and robots for autonomous driving and robot obstacle avoidance.
Lip sync and digital human presenters
Align a digital human's lip movements with speech for virtual hosts, newscasts and customer service.
Vision-language-action and grasping
Drive robotic grasping from vision and language instructions for embodied AI.
General writing and chat
Text generation, rewriting, Q&A and multi-turn chat for general assistant use cases.
Summarization and rewriting
Condense long text, rewrite in a new style and extract key points for reports, news and meeting notes.
Machine translation and localization
Translate between languages and polish for local audiences across documents, interfaces and subtitles.
Text content moderation
Flag violating, violent or extremist, sexual and spam text for communities and content platforms.
Text embeddings and semantic similarity
Encode text as vectors for semantic search, deduplication, clustering and similarity matching.
Entity and relation extraction
Extract people, organizations, events and their relationships from text to build knowledge graphs.
Sentiment and opinion analysis
Determine sentiment polarity and opinion in text for public opinion monitoring, reviews and customer service QA.
Intent recognition and slot filling
Understand user intent and extract key slots for dialogue systems and ticket routing.
Tabular classification and risk scoring
Classify tables and forms and output risk scores for risk control, compliance and review.
Field and form extraction
Extract structured fields from contracts, receipts and forms to automate data entry.
Text language identification
Automatically identify the language of text for routing, pre-translation and content distribution.
Language identification
Identify the language of speech or text for multilingual customer service and meeting systems.
PII/PHI detection and redaction
Detect and redact personal and health information to meet privacy compliance.
AI security detection and response
Identify risks such as prompt injection, jailbreaks and harmful content, and trigger response policies.
Causal inference and uplift modeling
Estimate causal effects and incremental gains of interventions for marketing, operations and pricing.
Graphs, anti-fraud and relationship risk
Mine fraud rings and linked risks from graphs for anti-fraud and risk control.
Text-to-image and reference-based generation
Generate images from text or reference images for creative work, design and marketing assets.
Inpainting, outpainting and controllable editing
Repaint regions, extend beyond the frame and edit with precise control for retouching, e-commerce and design.
Image classification and fine-grained recognition
Recognize image categories and fine-grained subclasses for quality inspection, archiving and identification.
OCR text recognition
Recognize text in images, including receipts, ID documents, documents and scene text.
Layout, table and formula parsing
Parse document layouts, table structures and math formulas for document digitization.
Face detection, recognition and liveness
Face detection, matching and liveness checks for access control, identity verification and security.
Pose, gesture and keypoints
Recognize body pose, hand gestures and keypoints for interaction, fitness and animation.
Medical image detection and segmentation
Detect and segment lesions and organs for diagnostic support and research.
Supervised defect detection
Detect surface and internal product defects from labeled samples for industrial quality inspection.
Visual measurement and alignment
Measure dimensions, position and alignment for precision manufacturing and assembly.
Open-vocabulary detection & segmentation
Detect and segment objects described in natural language, with no fixed classes.
Image & video segmentation
Semantic, instance and panoptic segmentation for matting, labeling and analysis.
3D reconstruction & digital twins
Reconstruct 3D scenes from images or point clouds for digital twins, mapping and simulation.
Depth estimation
Monocular and stereo depth estimation for autonomous driving, AR and robotics.
Super-resolution, denoising & deblurring
Image enhancement and restoration for sharper quality and old photo repair.
Text/image to 3D assets
Generate 3D models and assets from text or images for games, e-commerce and the metaverse.
Speech-to-text ASR
Transcribe speech into text for meetings, customer service and subtitles.
Text-to-speech TTS
Turn text into natural speech for announcements, audio content and accessibility.
Real-time voice agents
Real-time voice conversations that take business actions, for call centers and voice assistants.
Speech-to-speech translation
End-to-end speech translation that keeps voice and prosody, for cross-language communication.
Voice cloning & conversion
Clone a voice or convert speaking style for dubbing and personalized TTS.
Speaker diarization
Tell speakers apart for meeting notes and multi-speaker transcription.
Voiceprint recognition & verification
Confirm identity by voiceprint for identity checks and fraud prevention.
Wake word detection
Detect a chosen wake word for low-power voice assistant activation.
Voice activity detection VAD
Separate speech from non-speech to cut costs, segment audio and power real-time interaction.
Noise reduction & speech enhancement
Suppress noise and enhance voices for calls, recordings and meetings.
Echo cancellation & dereverberation
Remove echo and reverb for hands-free calls and meeting room audio.
Environmental sound recognition
Recognize sound events around you for security, industry and urban sensing.
Music & sound effect generation
Generate music and sound effects for content creation, games and short videos.
RAG knowledge Q&A
Retrieval-augmented generation that answers from your enterprise knowledge base with fewer hallucinations.
Document Q&A & comparison
Ask, compare and summarize across documents such as contracts, reports and papers.
Enterprise knowledge assistant
An all-in-one assistant for internal knowledge search, Q&A and writing.
Multimodal retrieval
Cross-modal search: find images or videos by text, or text by image.
Visual Q&A & scene understanding
Answer questions about images for accessibility, education and inspections.
Real-time multimodal assistant
An interactive assistant that handles speech, images and text in real time.
Tool-calling & workflow agents
Call external tools and APIs to orchestrate workflows and complete complex tasks.
GUI & browser agents
Operate graphical interfaces and browsers to finish tasks, for automation and RPA.
Coding & DevOps agents
Code generation, understanding and review, plus operations automation.
Code generation & understanding
Code completion, generation, explanation, refactoring and unit test generation.
Multi-agent orchestration for research & ops
Multiple agents working together on research, operations and analysis tasks.
Personalized ranking
Rank results by user behavior for recommendations, search and ads.
Candidate retrieval & similar items
Retrieve candidates and recommend similar items for e-commerce, content and social.
Semantic search & reranking
Semantic retrieval + reranking for more relevant search results.
Predictive maintenance & remaining useful life
Predict equipment failures and remaining useful life from multi-source data.
Streaming anomaly & change detection
Detect anomalies and sudden changes in streaming data in real time.
Time series forecasting
Forecast trends, seasonality and anomalies in time series data.
Scheduling, routing & resource optimization
Solve scheduling, routing and resource allocation optimization problems.
AMR navigation & dispatch
Mobile robot navigation, obstacle avoidance and fleet dispatch.
Drone & on-site autonomous inspection
Autonomous drone inspection, recognition and alerts.
Multi-sensor perception & fusion
Fuse cameras, radar, LiDAR and other sensors for perception.
IMU/radar/LiDAR fusion
Fuse inertial, radar and LiDAR data for localization and perception.
Visual-inertial SLAM
Visual + inertial odometry and mapping for robotics and AR.
Load & environmental forecasting
Forecast power load and environmental metrics for energy and city management.