How We Use AI
Transparency about the models powering Filmvision
Our Commitment
We believe filmmakers deserve to know how AI tools work and where the technology comes from. This page explains the models we use, their training data provenance, and our policies around generated content.
Image Generation Models
FLUX
Learn moreBlack Forest Labs
State-of-the-art image generation for storyboards and concept art. We support FLUX Pro, FLUX Dev, and specialized variants for different use cases.
Training data: Trained on licensed and permissively-licensed image datasets
Stable Diffusion
Learn moreStability AI
Versatile image generation with fine-tuned control for creative workflows
Training data: Trained on LAION and other publicly available datasets
Recraft
Learn moreRecraft AI
Design-focused image generation optimized for visual consistency
Training data: Proprietary training dataset
Video Generation Models
Kling
Learn moreKuaishou
Video generation for motion previz with strong motion coherence
Training data: Proprietary training dataset
Veo
Learn moreGoogle DeepMind
High-quality video generation with cinematic motion
Training data: Google's proprietary datasets
MiniMax
Learn moreMiniMax
Fast video generation optimized for creative iteration
Training data: Proprietary training dataset
LTX Video
Learn moreLightricks
Real-time video generation for rapid prototyping
Training data: Proprietary training dataset
Text & Analysis Models
Claude
Learn moreAnthropic
Primary model for screenplay analysis, narrative assistance, and complex reasoning
Training data: Trained on publicly available text with Constitutional AI
GPT-4o
Learn moreOpenAI
Multimodal understanding for image analysis and visual descriptions
Training data: Trained on publicly available text and licensed data
Llama
Learn moreMeta (via Groq)
Fast inference for real-time suggestions and lightweight tasks
Training data: Trained on publicly available datasets
Llama Guard
Learn moreMeta (via Groq)
Content safety and moderation for responsible AI usage
Training data: Fine-tuned on safety-focused datasets
Qwen
Learn moreAlibaba
Multilingual support and specialized creative writing capabilities
Training data: Proprietary training dataset
Your Content Rights
- You own all content you create with Filmvision
- Your screenplays are encrypted and never used for training
- Generated images/videos are yours to use commercially
- We do not claim rights to your creative work
Content Provenance (Coming Soon)
We're implementing C2PA content credentials to provide verifiable provenance for AI-generated assets. This will allow you to:
- Prove assets were AI-generated (for compliance/disclosure)
- Track generation parameters and timestamps
- Export with embedded provenance metadata
Our Position on Training Data
We recognize the film industry's concerns about AI models trained on copyrighted content without permission. We:
- Prefer models with transparent, licensed training data
- Monitor legal developments around AI training
- Will add "IP-safe" model options as they become available
- Never use your uploaded content to train any models
Zero Data Retention
We use Zero Data Retention (ZDR) agreements with all our AI providers. This means:
- Your prompts and generated content are not stored by AI providers
- Your data is never used to train or improve provider models
- Requests are processed and immediately discarded by providers
- No logs of your creative work exist outside Filmvision's encrypted infrastructure
Data Protection
- Screenplays encrypted at rest (AES-256)
- Storyboard data encrypted with user-specific keys
- No data shared with AI providers beyond generation requests
- Generation requests contain only the specific prompt, not full screenplay
- All data stored in ISO 27001:2022 certified infrastructure
Contact
Questions about our AI practices? Email us at ai@filmvision.ai