Multimodal / Featured system
PromptSilo
A generative-media product that evolved from prompt marketplace to continuity-aware, provider-orchestrated filmmaking workstation.
01 / Context
Context
PromptSilo began as a marketplace and creative Studio, then accumulated the difficult realities of AI filmmaking: providers with different contracts, browser-only tools, long-running video jobs, continuity between shots, durable media, and editing after generation.
02 / Problem
Problem
A useful filmmaking product cannot treat every provider like the same prompt box. It must understand native capabilities, authenticate browser workers safely, survive timeouts and partial success, retain results outside ephemeral provider URLs, and carry creative context into continuation and editing.
03 / System
System
Across V2, V3, the Pack Builder experiment, and V4, PromptSilo connects typed provider models, Hermes-backed asynchronous jobs, Codex and vision paths, authenticated CDP workers, durable object storage, generation sessions, timeline tools, and product-level ownership boundaries.
04 / Architecture
See the connective layer.
＋Architecture mode
System architecture / PromptSilo
Creative intent becomes a capability-aware plan, then moves through a direct provider or authenticated worker, a monitored job boundary, durable asset storage, and an editing or continuation workflow.
- Brief, references, and continuity state
- Planning or vision
- Provider or browser-worker adapter
- Authenticated broker and job lifecycle
- Validated durable asset
- Studio edit, continuation, or delivery
05 / Hard parts
Where the work actually was
- 01
Representing provider-specific image and video controls without presenting unavailable combinations as valid.
- 02
Turning authenticated browser-only creative tools into observable workers with retries, partial results, and durable outputs.
- 03
Preserving continuity and session state across generation, continuation, editing, and restoration.
- 04
Keeping product status precise across production foundations, beta code, prototypes, and locally validated expansion work.
06 / What changed or was learned
The field note
The product evolved from storing prompts and assets to orchestrating an end-to-end creative system. The durable advantage is not a single model call; it is the typed, observable connective tissue that makes varied providers and fragile browser tools usable inside a real workflow.
07 / Evidence
Proof
Safe public evidence, translated from tests and runtime validation rather than raw internal logs.
Locally validated image and video generation sessions with provider-specific capabilities, persistent results, continuation, and editing paths.
Broker and browser-worker paths use polling, retries, timeouts, partial results, and durable asset handoff rather than assuming a single synchronous API.
Restoration work treated deployed source, built output, durable media, and current schema state as separate surfaces to verify.
Built frame-aware editing, video continuation, music, timeline rendering, session persistence, and multiple editor information architectures.
08 / Stack
Technology in service of the system
- Next.js App Router
- React 19
- TypeScript
- Vite
- Express
- Prisma
- PostgreSQL
- R2
- FFmpeg
- Stripe
- Hermes broker jobs
- CDP browser workers
- Codex OAuth
- Gemini vision
- xAI video
09 / Evolution
One product family
- V2
Marketplace and Studio
React, Vite, Express, Prisma, and PostgreSQL product with multimodal generation, creative post-production, durable object storage, commerce, creator assets, and licensing.
Production foundation - V3
Agentic creative pipeline
Hermes broker jobs, Codex paths, authenticated ChatGPT and Dzine browser workers, prompt vision, retries, partial results, continuation, music, and timeline rendering. Storyboard planning remained feature-flagged and disabled by default in the sample configuration.
Beta - V3 UI
Build a Pack
An unfinished guided flow that productized domain-specific prompt engineering and strict JSON planning.
Experimental prototype - V4
Production shell and filmmaking Studio
Next.js and React production foundation with provider-aware capability modeling, persistent generation sessions, frame-accurate editing, secure sidecar boundaries, durable media, authentication, social features, capture tooling, and four editor directions.
Production foundation; Studio locally validated
10 / Status and boundaries