Canonical source: the canonical features document for this domain is
deep-dive/features.md, rendered in the docs center. This file is its operational companion (service topology, APIs, schemas, runtime detail) — reconciled by linking, not duplicated (§2, §13).
The Generative AI Factory -- Multi-Modal Content Generation at Scale
Isis is the generative AI engine that powers content creation across the Oshun ecosystem. It handles every type of AI-powered generation -- images, video, audio, 3D models, and text -- through a unified job submission system with GPU worker management, workflow versioning, output tracking with full lineage and provenance, and quality assurance. Isis provides the generation backbone that other domains (Yemaya, Hathor, Bellona) rely on for their creative production pipelines.
Claim verification source of truth:
docs/domains/isis/remediation-v2/ISIS_V2_DOCS_CLAIMS_RECONCILIATION.md([UNSUPPORTED]markers indicate claims not backed by current implementation evidence).
Table of Contents#
- Executive Summary
- Generation Capabilities
- Workflow System
- 3D Generation Pipeline
- Video Generation and Processing
- Image Generation
- Audio Generation
- AI Provider Ecosystem
- GPU Worker Management
- Output Management and Lineage
- Quality Assurance
- Cost Optimization
- Model Registry and Discovery
- Anomaly Detection and Security
- CLI Tools
- Client SDK
- Platform Availability
- Access and Pricing
- Integration Features
Executive Summary#
Isis operates as a generation-as-a-service platform. Users submit generation jobs through a REST API, CLI, or client SDK. Jobs are queued, routed to appropriate GPU workers, executed against the best available AI provider, and the results are stored with complete provenance and lineage tracking.
| Capability | What Isis Provides |
|---|---|
| Generation Types | 16 modes across image, video, 3D, audio, and text |
| AI Providers | 15+ integrated providers with automatic routing |
| Workflow System | Versioned, templated, composable generation workflows |
| Output Tracking | Full lineage graph with provenance for every output |
| Quality Assurance | Automated quality scoring, benchmarking, and regression |
| Cost Optimization | Budget enforcement, cost prediction, and provider routing |
| GPU Management | Distributed GPU workers with multi-GPU orchestration |
Key Numbers#
| Metric | Value |
|---|---|
| Generation Types | 16 modes |
| AI Providers | 15+ integrated |
| 3D Generation Providers | 6 specialized |
| Database Models | 12 (752-line Prisma schema) |
| API Services | 3 (Generation, Workflow, Output) |
| GPU Executors | 4 built-in types |
| CLI Commands | 7 command groups |
Generation Capabilities#
16 Generation Modes#
Isis supports a comprehensive set of generation types organized into five categories:
Image Generation#
| Mode | Description |
|---|---|
| TEXT_TO_IMAGE | Generate images from text prompts with full control over |
| model, dimensions, steps, guidance scale, sampler, and seed | |
| IMAGE_TO_IMAGE | Transform existing images using text prompts with |
| configurable strength and denoising | |
| UPSCALING | AI-powered image upscaling with multiple algorithms |
| INPAINTING | Fill in or modify selected regions of existing images |
Video Generation#
| Mode | Description |
|---|---|
| TEXT_TO_VIDEO | Generate video clips from text descriptions |
| IMAGE_TO_VIDEO | Animate a static image into video |
3D Generation#
| Mode | Description |
|---|---|
| TEXT_TO_3D | Generate 3D models from text descriptions |
| IMAGE_TO_3D | Reconstruct 3D models from one or more reference images |
| GAUSSIAN_SPLATTING | NeRF-based 3D scene reconstruction from photos |
| MESH_PROCESSING | Post-processing and optimization of 3D meshes |
| TEXTURE_UPSCALE | AI upscaling of 3D model textures |
| BLENDER_RENDER | Blender scene rendering via GPU workers |
Audio Generation#
| Mode | Description |
|---|---|
| TEXT_TO_AUDIO | Generate audio (sound effects, ambient) from descriptions |
| VOICE_SYNTHESIS | Text-to-speech with multiple voice providers |
| MUSIC_GENERATION | AI-composed music tracks with mood and genre control |
Job Submission#
Every generation request follows the same flow:
- Submit -- Submit a job with type, parameters, model, and priority
- Queue -- Job enters a Redis-backed queue with configurable priority (low, normal, high, urgent)
- Route -- Job is routed to an appropriate GPU worker based on generation type and worker capabilities
- Execute -- Worker processes the job using the specified AI provider
- Store -- Results are stored in S3-compatible storage with full metadata
- Notify -- Optional webhook callback on completion or failure
Job Management#
- Job Listing -- Browse all jobs with pagination, filtering by status and type
- Job Status -- Real-time status tracking (pending, queued, processing, completed, failed, cancelled)
- Job Cancellation -- Cancel running or queued jobs
- Job Output Retrieval -- Download completed job outputs and artifacts
- Progress Tracking -- Real-time progress updates during generation
- Webhook Notifications -- HTTP callback on job completion or failure
Workflow System#
Workflow Definitions#
Workflows define reusable, composable generation pipelines:
- Workflow Creation -- Define multi-step generation workflows with dependencies between steps
- Workflow Categories -- Organize workflows by category (image, video, 3D, audio, composite, utility)
- Workflow Tags -- Apply tags for discovery and organization
- Workflow Browsing -- List and search available workflows
Workflow Versioning#
Full version control for workflows:
- Version History -- Every workflow change creates a new version with complete configuration snapshots
- Version Comparison -- Compare different versions of a workflow
- Version Rollback -- Revert to any previous workflow version
- [UNSUPPORTED] Git-Like Branching -- Branch and merge workflow versions for experimental changes
- Changelog Tracking -- Automatic changelog generation between versions
Workflow Templates#
- Template Library -- Pre-built templates for common generation patterns
- Template Variables -- Parameterized templates with configurable inputs, default values, and validation
- Template Sharing -- Share and discover templates across the organization
- Template Starring -- Favorite frequently-used templates for quick access
ComfyUI Integration#
Deep integration with ComfyUI for visual workflow design:
- ComfyUI SDK -- WebSocket-based SDK for executing ComfyUI workflows programmatically
- Custom Nodes -- Oshun-specific custom ComfyUI nodes for generation tasks
- Workflow Import -- Import existing ComfyUI workflows into Isis
- RunComfy Integration -- Serverless ComfyUI execution via RunComfy for on-demand GPU access without managing infrastructure
- [UNSUPPORTED] Comfy Cloud Native -- Multi-cloud ComfyUI execution across different cloud providers
3D Generation Pipeline#
Multi-Provider 3D Generation#
Isis integrates six specialized 3D generation providers:
| Provider | Company | Strengths |
|---|---|---|
| Rodin | Hyperhuman | High-quality character and object generation |
| Meshy | Meshy | Fast text-to-3D and image-to-3D with texturing |
| Tripo | TripoSR | Single-image 3D reconstruction |
| Trellis | Microsoft | Research-grade 3D generation |
| Hunyuan | Tencent | Large-scale 3D model generation |
| 3DFY.AI | 3DFY | Production-quality 3D asset generation |
3D Model Post-Processing#
Comprehensive mesh processing pipeline:
- Mesh Optimization -- Reduce polygon count while preserving visual quality with configurable target reduction ratios
- Interior Face Removal -- Remove hidden interior faces that waste GPU resources during rendering
- Mesh Smoothing -- Apply smoothing algorithms with configurable iterations and strength
- Subdivision -- Increase mesh detail through subdivision with configurable levels
- UV Unwrapping -- Automatic UV unwrap generation for texture mapping
- Texture Baking -- Bake high-poly detail onto low-poly meshes
- LOD Generation -- Automatically generate Level of Detail meshes for runtime performance optimization
- Mesh Statistics -- Analyze vertex count, face count, edge count, and topology quality
Topology Verification#
Production-quality topology analysis:
- Quad-Mesh Analysis -- Verify quad/tri/ngon face distribution
- Pole Detection -- Identify and report topology poles that affect deformation quality
- Edge Flow Metrics -- Measure edge flow quality for animation-ready meshes
- Topology Presets -- Pre-configured quality standards for different use cases (game assets, film, CAD, 3D printing)
Auto-Rigging#
Automated skeletal rigging for generated 3D models:
- Skeleton Types -- Humanoid, quadruped, bird, fish, insect, serpent, and custom skeleton templates
- Bone Naming Conventions -- Support for Mixamo, Unreal, Unity, Blender, and custom naming standards
- Face Rig Configuration -- Automated facial rigging with configurable blend shapes and joint counts
- Control Rig Generation -- IK/FK control rig creation with pole targets and constraints
- Skin Weight Algorithms -- Multiple skinning methods (heat map, geodesic, voxelized, nearest bone)
- Rigging Presets -- Pre-configured rigging presets for different use cases
Texture Enhancement#
Advanced texture processing:
- Texture Upscaling -- AI upscaling of model textures with multiple algorithms (ESRGAN, Real-ESRGAN, BSRGAN)
- Seamless Tiling -- Convert textures to seamless tiles for repetitive surfaces
- Material ID Extraction -- Automatically extract and separate material regions from combined textures
- UDIM Layout -- Support for UDIM texture tile layouts for high-resolution production assets
- Batch Processing -- Process multiple textures in parallel
Gaussian Splatting#
NeRF and Gaussian splatting support:
- Scene Reconstruction -- Create 3D scenes from photo collections
- Mesh Extraction -- Convert Gaussian splatting results to traditional polygon meshes for use in game engines
Prompt Engineering for 3D#
Specialized prompt tools for 3D generation:
- Built-In Templates -- 10 category-specific prompt templates (humanoid, creature, weapon, prop, environment, vehicle, architecture, furniture, food, plant)
- Prompt Builder -- Interactive prompt builder with style, material, lighting, and detail modifiers
- Prompt Analyzer -- Analyze prompts for potential issues with severity ratings and suggestions
- Prompt Optimizer -- AI-powered prompt optimization for better generation results
- Prompt History -- Track and search prompt history with usage statistics
Reference Image Processing#
Prepare reference images for image-to-3D generation:
- Background Removal -- Automatic background removal for cleaner 3D reconstruction
- Multi-View Generation -- Generate multi-view reference sets from single images (front, back, left, right, top, bottom, three-quarter views)
- Image Preprocessing -- Resize, enhance, and normalize images for optimal generation results
- Processing Pipelines -- Chain preprocessing steps into reusable pipelines
Result Management#
Tools for evaluating and selecting generated 3D content:
- Preview Generation -- Generate preview renders from multiple angles with configurable quality and lighting
- Turntable Animations -- Generate rotating turntable previews for quick evaluation
- Side-by-Side Comparison -- Compare multiple generation results across quality dimensions (geometry, topology, texture, proportions, detail)
- Auto-Selection -- Automated best-result selection based on configurable quality criteria
- Selection Workflow -- Multi-stage selection process with nomination, review, and final selection
Video Generation and Processing#
AI Video Generation#
- Zeroscope -- Open-source text-to-video generation
- Runway -- Professional video generation with Gen-3 models
Video Post-Processing#
Comprehensive video processing with GPU acceleration:
NVIDIA RTX Video Processing#
- Video Super Resolution (VSR) -- AI upscaling with quality modes (fast, balanced, quality, ultra quality) and scale factors (1.5x, 2x, 4x)
- HDR Conversion -- Convert SDR video to HDR with support for HDR10, HDR10+, Dolby Vision, and HLG formats
- Color Space Conversion -- Transform between color spaces (BT.709, BT.2020, DCI-P3, sRGB)
RIFE Frame Interpolation#
- Frame Rate Doubling -- Smooth video by interpolating intermediate frames
- Multiple Models -- Support for RIFE v4.0 through v4.22 with quality/speed tradeoffs
- 4K Support -- Frame interpolation at up to 4K resolution
Topaz Video Enhancement#
- AI Upscaling -- Video upscaling with Topaz models (Proteus, Artemis, Gaia, Chronos, Iris, Nyx)
- Denoising -- AI video denoising and noise reduction
- Stabilization -- Video stabilization with configurable strength
- Frame Interpolation -- Topaz-powered frame interpolation
Production Workflows#
Pre-built video processing workflow templates:
- 4K Upscale Workflow -- Combined upscaling and enhancement for 4K output
- Frame Interpolation Workflow -- Smooth video with frame rate doubling
- HDR Conversion Workflow -- SDR to HDR conversion pipeline
- Restoration Workflow -- Video restoration and enhancement for archival content
- Streaming Workflow -- Optimized processing for streaming delivery
- Professional Workflow -- Full production pipeline with all enhancement stages
Image Generation#
Provider Integration#
Multiple image generation providers:
- Stability AI -- Stable Diffusion models for image generation
- Civitai Generation -- On-platform generation using community models
- DALL-E -- OpenAI image generation
- Replicate -- Open-source model hosting for various image models
Advanced Image Techniques#
AnimateDiff Lightning#
- Animation from Stills -- Create animations from static images using AnimateDiff Lightning models
ControlNet#
- Guided Generation -- Control image generation with structural guidance (edge maps, depth maps, pose, segmentation, line art, etc.)
IP-Adapter#
- Image Prompting -- Use reference images as prompts for style-consistent generation
- Style Transfer -- Transfer artistic styles between images
- Face Preservation -- Maintain facial identity across generated images
InstantID#
- Zero-Shot Identity -- Preserve identity in generated images without fine-tuning
Florence-2#
- Vision-Language Understanding -- Extract structured descriptions from images for guided regeneration and editing
Live Preview#
- Real-Time Preview Streaming -- See generation progress in real-time as images are being generated
- Interactive Editing -- Make adjustments during the generation process
Audio Generation#
Text-to-Speech#
- ElevenLabs -- High-quality voice synthesis with multiple voices and emotional tones
- Voice Cloning -- Clone existing voices for consistent narration
Text-to-Audio#
- Sound Effect Generation -- Generate sound effects from text descriptions
- Ambient Audio -- Create ambient soundscapes for scenes and environments
Music Generation#
- AI Music Composition -- Generate music tracks with mood, genre, tempo, and instrumentation controls
Conversational AI#
ElevenLabs Agents Platform 2.0 integration:
- Multi-Agent Voice -- Deploy multiple conversational AI agents with distinct voices
- Configurable LLM Backend -- Use GPT-4, Claude, or other LLMs as the conversation engine behind voice agents
AI Provider Ecosystem#
Unified Provider Architecture#
Isis abstracts away provider differences behind unified interfaces:
| Category | Providers |
|---|---|
| LLM | Anthropic (Claude), OpenAI (GPT), Google (Gemini), |
| xAI (Grok), Ollama (local) | |
| TTS | ElevenLabs |
| Image | Stability AI, Civitai, Replicate, OpenAI (DALL-E) |
| Video | Zeroscope, Runway |
| Video Processing | NVIDIA RTX, RIFE, Topaz |
| 3D | Rodin, Meshy, Tripo, Trellis, Hunyuan, 3DFY.AI |
| Model Registry | Civitai |
| Execution | RunComfy (serverless ComfyUI) |
Provider Factory#
- Automatic Instantiation -- Create providers from configuration with a single factory call
- Instance Caching -- Reuse provider instances across requests for efficiency
- Unified Error Handling -- Consistent error types across all providers
Intelligent Routing#
The Unified Generation Pipeline routes requests to the best available provider:
- Capability Matching -- Match request requirements to provider capabilities
- Quality Tier Selection -- Route based on desired quality tier (draft, standard, premium, ultra)
- Fallback Chains -- Automatic fallback to alternative providers on failure
- Routing Strategies -- Cost-optimized, quality-optimized, latency- optimized, and balanced routing
GPU Worker Management#
Worker Architecture#
- Redis Queue Polling -- Workers poll Redis queues for pending jobs at configurable intervals
- Multi-Executor Support -- Four built-in executors: texture-upscale, blender-render, gaussian-splatting, mesh-processing
- Worker Registration -- Workers register their GPU capabilities (model, VRAM, compute capability) for intelligent job routing
- Graceful Shutdown -- Clean shutdown on SIGINT/SIGTERM with job completion or requeue
Multi-GPU Orchestration#
- Distributed Execution -- Distribute generation jobs across multiple GPUs
- Pipeline Parallelism -- Split large jobs across multiple GPUs for faster completion
- Hardware Detection -- Automatic detection of available GPU hardware and capabilities
- Memory Optimization -- Configurable memory optimization for different GPU tiers (consumer, professional, datacenter)
- VRAM Estimation -- Estimate VRAM requirements before job submission to prevent out-of-memory failures
Output Management and Lineage#
Output Manifest System#
Every generated output is tracked with a comprehensive manifest:
- File Metadata -- Format, dimensions, duration, file size, checksums (MD5, SHA-256)
- Quality Tier -- Draft, standard, high, production, or master quality classification
- License Information -- License type (CC-BY, CC-BY-SA, CC-BY-NC, CC0, MIT, Apache-2.0, proprietary, custom) with usage rights specification
- Manifest Validation -- Automated validation of manifest completeness and correctness
Provenance Tracking#
Complete generation history for every output:
- Generator Information -- Which model, provider, and version produced the output
- Input Parameters -- All prompts, settings, seeds, and configuration used
- Prompt Extraction -- Extract and store the original prompt from provenance records
- Provenance Summary -- Human-readable summaries of generation history
Lineage Graph#
Track how outputs relate to and derive from each other:
- Lineage Nodes -- Every output is a node in the lineage graph
- Relationship Types -- Derived from, variant of, upscale of, inpaint of, composite of, reference for, refined from, and batch sibling relationships
- Graph Visualization -- Visual representation of lineage relationships
- Graph Serialization -- Export and import lineage graphs for archival and sharing
- Graph Statistics -- Analyze lineage depth, breadth, and relationship distribution
Retention Policies#
Configurable output lifecycle management:
- Retention Rules -- Define how long outputs are retained based on quality tier, age, and usage
- Storage Tiering -- Automatic migration between hot, warm, and cold storage tiers based on access patterns
- Policy Management -- Create, list, and configure retention policies via API
Quality Assurance#
Automated Quality Scoring#
- Multi-Dimensional Scoring -- Score outputs across quality dimensions (sharpness, coherence, prompt alignment, artifact presence, composition)
- Quality Profiles -- Pre-configured quality standards for different use cases
- Quality Tier Determination -- Automatically classify outputs into quality tiers based on scores
Artifact Detection#
- Visual Artifacts -- Detect common generation artifacts (blurriness, distortion, color banding, repetition patterns)
- Severity Classification -- Rate artifact severity for automatic accept/reject decisions
Benchmarking#
- Test Suites -- Standardized test prompts for consistent quality comparison across models and providers
- Benchmark Runs -- Execute benchmark suites and collect performance data
- Regression Detection -- Automatically detect quality regressions between model versions or provider updates
- Comparative Analysis -- Side-by-side comparison of results across providers
User Feedback Loop#
- Feedback Collection -- Collect user ratings and feedback on generated outputs
- Feedback Aggregation -- Aggregate feedback to identify quality trends and improvement opportunities
Cost Optimization#
Budget Management#
- Budget Enforcement -- Set spending limits per user, team, project, or organization
- Budget Periods -- Daily, weekly, monthly, quarterly, and annual budget periods
- Budget Alerts -- Configurable alerts at threshold levels (50%, 75%, 90%, 100%) with multiple severity levels
- Budget Restrictions -- Automatically restrict access when budgets are exceeded
Cost Tracking#
- Per-Request Cost -- Track generation cost per request with provider- specific pricing
- Cost Attribution -- Attribute costs to users, teams, projects, and categories
- Cost Aggregation -- Aggregate costs by time period, provider, model, generation type, and quality tier
- Multi-Currency -- Support for multiple currencies with exchange rate conversion
Cost Optimization Engine#
- Optimization Strategies -- Cost-optimized, quality-first, balanced, speed-first, and budget-constrained strategies
- Batch Optimization -- Optimize costs for batch generation requests by routing to cost-effective providers
- Caching Strategies -- Reuse cached results for identical or similar requests
- Cost Forecasting -- Predict future costs based on historical usage patterns
- Efficiency Metrics -- Track cost-per-quality-point for optimization decisions
- Quota Management -- Usage quotas with configurable limits and overage policies
Analytics and Reporting#
- Custom Dashboards -- Configure analytics dashboards with multiple widget types (chart, counter, table, heatmap, gauge, progress, timeline, map)
- Report Templates -- Pre-built report templates for cost, quality, and usage analysis
- Scheduled Reports -- Automated report generation and delivery
- SLA Monitoring -- Define and monitor service level agreements with compliance tracking
- Anomaly Detection -- Automatic detection of unusual cost or usage patterns
Model Registry and Discovery#
Civitai Integration#
Deep integration with Civitai, the largest AI model community:
- Model Discovery -- Search and browse models by type, architecture, and popularity
- Model Download -- Download model checkpoints, LoRAs, embeddings, and other model files
- Model Versioning -- Track model versions with metrics and changelogs
- Popular Model Catalog -- Quick access to popular models across categories
LoRA Training#
- Custom LoRA Training -- Train custom LoRA models on user-provided images
- Training Configuration -- Configure training parameters, dataset preparation, and quality targets
- Training Progress -- Monitor training progress and intermediate results
Model Merging#
- Model Combination -- Merge multiple models with configurable blend ratios and methods
Model Comparison#
Compare models head-to-head:
- Standard Test Prompts -- Evaluate models against standardized prompt suites
- Quality Metrics -- Compare generation quality, speed, memory usage, and consistency
- Statistical Analysis -- Rigorous statistical comparison with significance testing
- Recommendation Engine -- AI-generated recommendations based on use case requirements
Model Quality Assessment#
- Rating System -- Community ratings with weighted scoring and confidence calculation
- Review System -- Detailed reviews with sentiment analysis
- Download and Generation Metrics -- Track model popularity and usage
- FID/CLIP Scores -- Automated quality assessment using standard image quality metrics
- Leaderboard Rankings -- Model rankings by category with percentile scores
- Trend Analysis -- Track model quality trends over time
Collection Management#
- Model Collections -- Curate collections of models for different use cases and styles
- Smart Curation -- Automated collection curation based on quality criteria
- Collection Sync -- Keep collections synchronized with latest model versions
- Sharing -- Share collections with team members or publicly
Content Safety#
- NSFW Detection -- Automated content safety classification
- Safety Scoring -- Multi-dimensional safety scoring for generated content
- Policy Enforcement -- Configurable safety policies per organization
Resource Recommendations#
- Use-Case Recommendations -- AI-powered model recommendations based on intended use case, quality requirements, and performance constraints
Creator Monetization#
- Buzz Currency -- Civitai's Buzz currency integration for model usage payments
- Cost Estimation -- Estimate generation costs before submission
- Creator Payouts -- Revenue tracking for model creators
Civitai Link#
- Local ComfyUI Bridge -- Connect local ComfyUI installations to Civitai for seamless model management and generation
Anomaly Detection and Security#
Suspicious Activity Detection#
- Usage Pattern Analysis -- Monitor user behavior patterns for anomalies
- Activity Thresholds -- Configurable thresholds for generation frequency, API calls, and resource consumption
- Risk Scoring -- Multi-factor risk level assessment (low, medium, high, critical)
- Automated Actions -- Configurable responses (alert, throttle, restrict, suspend, ban) based on risk level
Chargeback Prevention#
- Chargeback Prediction -- AI-powered prediction of potential chargeback events
- Feature-Weighted Scoring -- Configurable feature weights for chargeback risk factors
- Transaction Monitoring -- Real-time monitoring of payment transactions
Account Protection#
- Login Pattern Analysis -- Monitor login behavior for compromise indicators
- Device Fingerprinting -- Track device patterns for session validation
- Geographic Analysis -- Monitor geographic access patterns for anomalies
- Session Management -- Automatic session invalidation on suspicious activity
CLI Tools#
Command Groups#
The Isis CLI provides seven command groups for complete platform interaction:
| Command | Description |
|---|---|
generate |
Submit generation jobs (text-to-image, image-to-image, |
| text-to-video, image-to-video, text-to-3D, image-to-3D, | |
| text-to-audio, voice-synthesis, music, upscale, inpaint) | |
jobs |
List, view status, cancel, and download job outputs |
workflows |
Create, list, view, and manage generation workflows |
outputs |
List, view, and manage generated outputs |
config |
Configure CLI settings, API endpoint, and credentials |
health |
Check API service health and readiness |
auth |
Authenticate and manage CLI sessions |
Interactive Mode#
- Interactive Generation -- Step-by-step guided generation with prompt input, model selection, and parameter configuration
- Progress Tracking -- Real-time progress bars and spinners during generation
- Automatic Download -- Optionally download completed results immediately
Developer Experience#
- Aliases -- Short aliases for common commands (e.g.,
genforgenerate,t2ifortext-to-image) - JSON Output -- Machine-readable JSON output for scripting and automation
- Webhook Support -- Configure webhook URLs for completion notifications
Client SDK#
TypeScript/JavaScript SDK#
The @isis/client SDK provides full-featured typed access:
- Job Submission -- Submit any generation type with typed request objects
- Job Management -- List, status check, cancel, and retrieve outputs
- Workflow Management -- CRUD operations for workflows and templates
- Output Management -- List and retrieve output manifests
- Health Checks -- API health and readiness monitoring
- Typed Errors -- Specific error types for different failure modes
- HTTP Transport -- Configurable HTTP client with retry and timeout
Platform Availability#
| Platform | Port | Description |
|---|---|---|
| Generation API | 4001 | REST API for job submission, model registry, health |
| Workflow Registry | 3001 | REST API for workflow and template management |
| Output Registry | 3002 | REST API for output manifests and retention policies |
| GPU Worker | -- | Background service polling Redis for jobs |
| CLI | -- | Command-line interface for all operations |
| TypeScript SDK | -- | npm package for programmatic access |
Access and Pricing#
Generation Credits#
| Tier | Credits/Month | Priority | GPU Access |
|---|---|---|---|
| Free | 100 | Low | Shared queue |
| Starter | 1,000 | Normal | Shared queue |
| Professional | 10,000 | High | Priority queue |
| Enterprise | Custom | Urgent | Dedicated GPUs |
Priority Levels#
| Priority | Queue Behavior |
|---|---|
| Low | Processed after all higher-priority jobs; may be preempted |
| Normal | Standard FIFO processing within the normal queue |
| High | Processed ahead of low and normal priority jobs |
| Urgent | Immediate processing; may interrupt lower-priority jobs |
Integration Features#
Cross-Domain Events#
Isis publishes lifecycle events consumed by other domains:
| Event | Consumers | Purpose |
|---|---|---|
isis.job.submitted |
Yemaya, Hathor | Job tracking |
isis.job.completed |
Yemaya, Hathor, Bellona | Result delivery |
isis.job.failed |
Yemaya, Hathor | Error handling |
isis.asset.generated |
Bellona | Engine import trigger |
isis.output.stored |
Yemaya | Asset library update |
Events Consumed#
| Source Event | Purpose |
|---|---|
yemaya.generation.requested |
Creative studio generation requests |
hathor.asset.requested |
Worldbuilding asset generation |
Infrastructure Integration#
| Service | Purpose |
|---|---|
| PostgreSQL | Job queue, workflows, outputs, provenance, audit log |
| Redis | Job queuing, worker coordination, caching |
| MinIO / AWS S3 | Generated file storage (hot/warm/cold tiering) |
| Nx Build | Monorepo build, test, lint, and serve orchestration |