Last updated: March 2026
Internal register maintained pursuant to the EU AI Act (Regulation (EU) 2024/1689). Published for transparency and accountability.
This register documents all AI systems deployed within the Katoads platform, including their risk classification, purpose, data processing details, and responsible parties. It is maintained by the Katoads compliance team and reviewed quarterly.
| Provider | Google (Gemini API) |
| Model / Version | gemini-3.1-flash-lite-preview |
| Purpose & Deployment Context | Analyzes product information (images, descriptions, URLs) to extract structured marketing data: target audiences, key selling points, benefits, psychological leverages, and suggested visual cues for ad composition. |
| AI Act Risk Level | Limited risk — Content generation for commercial advertising. No high-risk use case per Annex III. |
| Input Data | Product images (uploaded or via URL), product name, description, price, category, brand information, target language. |
| Output Data | Structured analysis JSON: marketing hooks, CTAs, target audience profiles, psychological leverages, visual cue suggestions. |
| Human Oversight Measures | Users review analysis results. Analysis can be regenerated. Output feeds into generation pipeline but user controls all final parameters. |
| Last Risk Assessment | Q1 2026 |
| Provider | Google (Gemini API) |
| Model / Version | gemini-3.1-flash-image-preview |
| Purpose & Deployment Context | Generates advertising images from text prompts. Prompts incorporate product context, marketing leverages, composition instructions, and style parameters to produce ad-ready visuals. |
| AI Act Risk Level | Limited risk — Synthetic content generation. Article 50(2) transparency obligations apply. |
| Input Data | Text prompt (generated from product analysis), style parameters, platform format, aspect ratio. |
| Output Data | Generated advertising image (PNG, up to 1024x1024). Stored in Vercel Blob. Metadata includes model version and generation parameters. |
| Human Oversight Measures | All generated images are presented for user review. Users may regenerate, edit via layout editor, or discard. No automatic publication to ad platforms. |
| Last Risk Assessment | Q1 2026 |
| Provider | fal.ai |
| Model / Version | Kling 2.6 Pro |
| Purpose & Deployment Context | Generates short-form video ads from static images for social media platforms (TikTok, Instagram Reels, YouTube Shorts). |
| AI Act Risk Level | Limited risk — Synthetic video content generation. Article 50(2) transparency obligations apply. |
| Input Data | Source image (generated or uploaded), video duration, aspect ratio. |
| Output Data | Generated video file (MP4). Stored in Vercel Blob. |
| Human Oversight Measures | Users review generated videos before download or export. All videos are clearly labeled as AI-generated within the platform. |
| Last Risk Assessment | Q1 2026 |
| Provider | Google (Gemini API) |
| Model / Version | gemini-3.1-flash-lite-preview |
| Purpose & Deployment Context | Generates multilingual marketing copy (headlines, CTAs, benefit statements) and optimized image generation prompts based on product analysis results. |
| AI Act Risk Level | Limited risk — Text content generation for advertising purposes. |
| Input Data | Product analysis data, target language, psychological leverage selection, style preferences, brand template data. |
| Output Data | Marketing copy strings (headlines, subheadlines, CTAs) in the target language, image generation prompts. |
| Human Oversight Measures | Generated copy is displayed for user review. Users can edit text via the layout editor. Quality filters automatically flag language-specific banned phrases and generic copy. |
| Last Risk Assessment | Q1 2026 |
© 2026 Katoads. All rights reserved.