Our AI Models

FOTOhub develops proprietary AI systems that power core platform capabilities. These models run on our dedicated GPU infrastructure across European data centers, providing low-latency inference, full data sovereignty, and cost advantages over cloud-only providers.

Gabriel AI Orchestrator

Gabriel is FOTOhub's intelligent platform orchestrator — a proprietary AI layer that sits between users and 90+ AI models, making the entire platform feel like a single coherent assistant.

What Gabriel Does

Capability	Description	Latency
Intent Classification	Understands natural language requests and maps them to platform features	<100ms
Model Selection	Dynamically selects the optimal model based on task, quality needs, and cost	<50ms
Prompt Engineering	Enriches user prompts with model-specific optimizations	<200ms
Multi-step Workflows	Orchestrates complex tasks requiring sequential operations	<500ms
Proactive Suggestions	Context-aware recommendations as users navigate the platform	<50ms
Auto-translation	Real-time translation across 30+ languages	<300ms

Architecture

User Input → Gabriel Orchestrator
                 ├── Intent Classifier (proprietary fine-tuned model)
                 ├── Model Router (availability + cost + quality scoring)
                 ├── Prompt Enhancer (per-model architecture knowledge)
                 ├── Workflow Engine (multi-step task decomposition)
                 └── Feedback Loop (learns from user corrections)

Benchmarks

Metric	Score	Details
Intent accuracy	96.2%	Tested on 10,000+ real user queries across PL/EN/DE
Routing precision	98.7%	Correct model selected for the task type
Prompt enhancement lift	+34%	Quality improvement measured by user satisfaction ratings
Response latency (P95)	180ms	Single-shot classification, EU datacenter
Streaming first token	<80ms	SSE stream begins before full classification completes
Availability	99.97%	Multi-model failover chain, auto-recovery

Key Features

• Dynamic Model Awareness — Only recommends models that are currently healthy (checked every 60s)
• Cost Optimization — Routes to the best quality-per-credit model for each task
• 10 Function-Calling Tools — Image, Video, Chat, Music, 3D, Editing, Brand, Tools, Workflows, Q&A
• Multi-language — Natively handles PL, EN, DE, FR, ES + 25 more
• Zero Extra Cost — Gabriel routing is free; you only pay for the downstream generation

API Access

POST https://apis.fotohub.app/v1/ai/gabriel         # Single-shot classification
POST https://apis.fotohub.app/v1/ai/gabriel/stream   # Streaming SSE
POST https://apis.fotohub.app/v1/ai/gabriel/suggest  # Autocomplete (no auth)
POST https://apis.fotohub.app/v1/ai/gabriel/recommend # Proactive tips (no auth)

See Gabriel AI API Reference for complete endpoint documentation.

---

IDA Music

IDA Music is FOTOhub's proprietary music generation engine. It produces full songs with vocals, lyrics, and complex multi-instrument arrangements up to 8 minutes long — running entirely on our GPU infrastructure.

Capabilities

Feature	Specification
Duration	30 seconds to 8 minutes
Output	Full songs with vocals OR instrumentals
Vocals	Male/female, 12+ languages (EN, ZH, JA, KO, ES, FR, DE, IT, PT, RU, AR, HI)
Genres	Any genre — pop, rock, electronic, jazz, classical, hip-hop, ambient, cinematic, folk, metal
Lyrics	Custom lyrics with structure tags, or AI-generated
Quality Presets	Draft (15s), Standard (30s), High (45s), Max (90s)
Formats	MP3, WAV, FLAC
Batch	1–4 variations per request
Seed Control	Reproducible outputs

Audio Quality Benchmarks

Metric	IDA Music	Industry Average
Sample rate	48 kHz	44.1 kHz
Bit depth	24-bit	16-bit
Frequency response	20 Hz – 20 kHz	20 Hz – 16 kHz
Dynamic range	96 dB	72 dB
THD+N	<0.01%	<0.05%
Vocal clarity (PESQ)	4.2 / 5.0	3.6 / 5.0
Musical coherence	4.5 / 5.0	3.8 / 5.0
Lyric intelligibility	94%	82%

Quality Presets

Preset	Steps	Guidance	Sampler	Generation Time
Draft	20	5.0	Euler	~15 seconds
Standard	40	5.0	Euler	~30 seconds
High	50	7.0	Euler	~45 seconds
Max	50	7.0	Heun + ADG	~90 seconds

Song Structure Control

IDA Music supports detailed structural tags for professional music production:

[intro], [verse], [chorus], [bridge], [outro],
[pre-chorus], [hook], [interlude], [break],
[drop], [solo], [ad-lib], [fade-out]

Musical Parameters

Parameter	Range	Description
BPM	60–220	Tempo control
Key/Scale	All major/minor keys	Tonal center
Time Signature	4/4, 3/4, 6/8, 5/4, 7/8	Rhythmic structure
Vocal Language	12+ languages	Singing language
Vocal Gender	Male / Female / Auto	Voice type

Pricing

Duration	Credits	PLN
≤ 3 minutes	2	0.15
> 3 minutes	4	0.30

Cost Advantage

IDA Music costs 2–4 credits per generation. Comparable cloud services charge 5–25 credits for similar output. Running on our own infrastructure means: lower latency for European users, no external rate limits, and full control over output quality.

API Access

POST https://apis.fotohub.app/v1/ai/generate/music          # Generate music
POST https://apis.fotohub.app/v1/ai/generate/music/compose   # AI lyrics composer (free)

See Music & Audio API Reference for complete endpoint documentation.

---

IDA Voice

FOTOhub's proprietary voice synthesis system providing natural-sounding speech with voice cloning capabilities.

Capabilities

Feature	Specification
Languages	30+ languages with native pronunciation
Voice Cloning	Custom voice from audio sample
Expressiveness	Emotion, pace, emphasis control
SSML Support	Fine-grained prosody control
Streaming	Real-time audio streaming for low latency
Max Input	5,000 characters per request

Voice Quality

Metric	IDA Voice	Standard TTS
MOS (Mean Opinion Score)	4.3 / 5.0	3.7 / 5.0
Naturalness	4.4 / 5.0	3.5 / 5.0
Intelligibility	98%	95%
Emotion accuracy	89%	71%
Clone similarity	92%	—

Pricing

Model	Credits	Per
IDA Voice	2	1,000 characters

API Access

POST https://apis.fotohub.app/v1/ai/generate/speech

---

IDA Audio (Sound Effects)

Instant generation of sound effects and ambient audio from text descriptions.

Capabilities

Feature	Specification
Duration	1–30 seconds
Categories	Nature, mechanical, sci-fi, UI, foley, ambient, impacts, musical
Format	MP3 (default), WAV
Latency	2–5 seconds generation time

Pricing

Credits	Per
3	generation (any duration)

API Access

POST https://apis.fotohub.app/v1/ai/generate/sfx

---

Platform-wide AI Features

Beyond dedicated models, FOTOhub's AI layer provides:

Feature	Technology	Cost
Smart Search	Semantic vector search across user's media library	Free
Auto-tagging	Automatic metadata extraction from uploaded images	Free
Brand Consistency	AI-enforced style matching from brand kits	Free
Content Safety	Real-time NSFW and policy detection	Free
Prompt Translation	Automatic prompt translation to optimal model language	Free

---

Infrastructure

All proprietary models run on FOTOhub's dedicated GPU cluster:

Resource	Specification
Location	EU (Frankfurt, eu-central-1)
GPU	NVIDIA A100 80GB / L4
Inference	Custom optimized pipeline
Availability	99.97% uptime (multi-node redundancy)
Data	GDPR-compliant, no training on user data
Latency	<200ms P95 for European users

Why Self-Hosted

Advantage	Benefit
Cost	60–80% lower per-inference cost vs cloud APIs
Latency	Direct EU routing, no transcontinental hops
Privacy	Data never leaves FOTOhub infrastructure
Control	Custom quality tuning, no provider deprecation risk
Rate Limits	No external throttling — capacity scales with demand

---

Comparison

FOTOhub vs. Cloud-Only Platforms

Capability	FOTOhub	Cloud-Only
Music generation (8 min)	4 credits (~$0.30)	$2–5 per track
Voice synthesis	2 credits/1K chars	$0.015–0.030/1K chars
Intelligent routing	Free (Gabriel)	Not available
Model selection	Automatic, 90+ models	Manual, 1 provider
Prompt enhancement	Free, per-model optimized	Not available
Multi-step workflows	Orchestrated by Gabriel	Manual chaining
EU data residency	Default	Extra cost / unavailable

Continuous Improvement

Our proprietary models are continuously improved through architecture upgrades, training data curation, and inference optimization. Updates are deployed seamlessly — no API changes required.