Aeion AI for Content Generation
Image, video, voice, 3D mesh, document parsing — all through one engine, one API, one cost-tracking surface. Creative-modality adapters (image, voice TTS/STT, video, 3D) plus LLM adapters for prompt engineering, all part of the same 57-provider registry. Track spend per modality via `ai_usage_logs`.
The Content-Generation Stack
A modern AI creative product typically touches 4-6 modalities. Aeion's engine routes through one API regardless of provider.
Per-Modality Cost Realities
Cost varies wildly across modalities. Plan accordingly.
Per-Modality Spend Visibility (and Where You Still Have To Build)
ai_budgets today is a single tenant-wide cap — it doesn't have a per-modality dimension, so you can't set "warn at $100 on LLM but hard-block at $50 on video" as separate budget records. What you DO get for free is per-modality visibility: UsageService.getUsageByDimension / getAggregatedUsage breaks down ai_usage_logs by modality (text / image / audio / video / embedding), so you can see exactly where spend is concentrated.
Provider Routing for Content
Different content types route to different providers automatically.
FAQ
Aeion's Redis-backed job queue handles async modalities. For image / video / 3D generation, queue the request, return a job ID immediately, and let the user poll or subscribe to the completion event.
Yes — per-tenant or per-user provider configs override the platform defaults, on every module, no upsell required. Users can provide their own keys for any of the 57 providers.
Currently Aeion ships text-to-music via 3rd party API providers; native music-gen adapter is on Q4 2026 roadmap. For voice music + sound effects, ElevenLabs Sound Effects API works through the standard TTS adapter.
Cartesia Sonic + Deepgram streaming STT + Groq Llama 3.1 70B = end-to-end p50 ≤ 700ms for voice-to-voice interactions. See the `voice-products` use-case for the full architecture.
Stability has built-in safety filters; configure threshold per-tenant. Pair with OpenAI's free moderation API as a pre-filter on the prompt itself.