Intro & scope
Use cases where AI creates or transforms media and design artefacts. Patterns: GEN.IMAGE (marketing imagery, editing, restoration, headshots, comics, synthetic training images), GEN.VIDEO (video generation, avatars, automated editing, personalised video, film/VFX), GEN.AUDIO (music and sound effects) and GEN.DESIGN (graphics, slides, logos, 3D, architecture, game content, product design, motion capture, site planning, websites).
Scope boundaries: - GEN-001 (campaign and product imagery from scratch) vs GEN-011 (editing existing photos) vs GEN-018 (room staging) vs GEN-026 (identity-preserving portraits). - GEN-002 (generative clips) vs GEN-012 (editing existing footage) vs GEN-003 (avatar presenter) vs GEN-015 (template-plus-data personalised video) vs GEN-021 (professional film/VFX pipelines). - GEN-008 (generated 3D assets) vs GEN-025 (reconstruction of real scenes by capture) vs GEN-022 (motion capture and animation). - GEN-009 (architecture visualisation) vs GEN-024 (generative layouts and massing with analysis) vs GEN-018 (staging of existing listings). - GEN-004 (layouts and banners) vs GEN-005 (slides) vs GEN-006 (logos and brand kits) vs GEN-027 (full websites). - GEN-020 generates synthetic training data for VIS models, not marketing content.
AI Act: nearly all are limited risk (Art. 50: label synthetic images, audio and video; deepfake disclosure for avatars, try-on and portraits). GEN-020, GEN-024, GEN-025 and GEN-027 are minimal risk. None is high-risk by default. GDPR: personal data is certain for avatars, headshots, try-on, motion capture and personalised video (consent, likeness rights, voice cloning), otherwise depends on inputs. Copyright and licence clarity of models and training data is the main non-AI-Act risk.
Use cases
| ID | Title | Pattern | Maturity | Path | HW tier | AI Act risk |
|---|---|---|---|---|---|---|
| GEN-001 | Marketing imagery & AI product photography | GEN.IMAGE | mainstream | A | HW2 | limited |
| GEN-002 | Video generation for marketing & social media | GEN.VIDEO | early-adopter | A | HW3 | limited |
| GEN-003 | AI avatar / presenter videos for training & communication | GEN.VIDEO | mainstream | A | HW3 | limited |
| GEN-004 | Graphic design & layout generation (banners, flyers) | GEN.DESIGN | mainstream | A | HW2 | limited |
| GEN-005 | Presentation & slide generation | GEN.DESIGN | mainstream | A | HW2 | limited |
| GEN-006 | Logo & brand asset generation | GEN.DESIGN | mainstream | A | HW2 | limited |
| GEN-007 | Music & sound-effect generation | GEN.AUDIO | early-adopter | A | HW2 | limited |
| GEN-008 | 3D asset generation for games, AR & e-commerce | GEN.DESIGN | early-adopter | A | HW3 | limited |
| GEN-009 | Architecture & interior design visualisation | GEN.DESIGN | early-adopter | A | HW2 | limited |
| GEN-010 | Fashion design & virtual try-on | GEN.IMAGE | early-adopter | A | HW3 | limited |
| GEN-011 | Image editing & retouching (background removal, upscaling, inpainting) | GEN.IMAGE | commodity | A | HW2 | limited |
| GEN-012 | Automated video editing (cutting, reframing, highlight clips) | GEN.VIDEO | mainstream | A | HW2 | limited |
| GEN-013 | Game content generation (NPC dialogue, levels, assets) | GEN.DESIGN | early-adopter | B | HW2 | limited |
| GEN-014 | Product & industrial design ideation | GEN.DESIGN | early-adopter | A | HW2 | limited |
| GEN-015 | Personalised video at scale | GEN.VIDEO | mainstream | A | HW3 | limited |
| GEN-016 | Illustrated stories, comics & children's content | GEN.IMAGE | early-adopter | B | HW2 | limited |
| GEN-017 | Photo restoration & colourisation | GEN.IMAGE | mainstream | A | HW2 | limited |
| GEN-018 | Virtual staging & real-estate visualisation | GEN.IMAGE | mainstream | A | HW2 | limited |
| GEN-020 | Synthetic image data generation for vision model training (rare defects, edge cases) | GEN.IMAGE | early-adopter | B | HW2 | minimal |
| GEN-021 | Film, TV & ad production: concept art, storyboards, previs & generative VFX | GEN.VIDEO | early-adopter | B | HW4 | limited |
| GEN-022 | AI character animation & markerless motion capture from video | GEN.DESIGN | early-adopter | A | HW2 | limited |
| GEN-024 | Generative site planning, massing & floor-plan layout | GEN.DESIGN | early-adopter | A | HW1 | minimal |
| GEN-025 | Photo/video-to-3D capture (Gaussian splatting, NeRF) for tours, products & digital twins | GEN.DESIGN | early-adopter | A | HW2 | minimal |
| GEN-026 | AI professional headshots & identity-preserving portrait generation | GEN.IMAGE | mainstream | A | HW2 | limited |
| GEN-027 | Website & landing-page generation (structure, copy, imagery) | GEN.DESIGN | mainstream | A | HW2 | minimal |
Shared stack
Common core (all tiers): ComfyUI or a comparable node-based workflow runner behind an API, a model registry with licence metadata, an asset store (S3-compatible, e.g. MinIO) with a DAM/brand-kit layer, a review-and-approval step, C2PA content credentials plus invisible watermarking on outputs (AI Act Art. 50), prompt/output safety filters (NSFW, brand, likeness checks) and a job queue (Celery/Temporal or n8n) for batch variants.
HW1 (CPU / small server). Layout and rule engines, OR-Tools/CP-SAT and surrogate models for GEN-024, templating (HTML/SVG/PPTX) driven by a hosted or 3-8B quantised LLM, rembg/ONNX background removal, ffmpeg + Whisper-small for simple cutting. Cloud API (EU region) for any diffusion step.
HW2 (single GPU 24-48 GB). HW1 plus FLUX/SDXL-class diffusion with ControlNet, IP-Adapter and LoRA training via ComfyUI/diffusers (GEN-001, 004, 006, 009, 011, 016, 017, 018, 026), SAM 2 segmentation, Real-ESRGAN/SUPIR upscaling, open VLM (Qwen-VL class) for checks and captions, ACE-Step/MusicGen for GEN-007, Whisper large + diarization and reframing with tracking for GEN-012, template-driven avatar/TTS rendering for GEN-015, Gaussian-splatting training for GEN-025, markerless mocap models for GEN-022, a 7-32B LLM via vLLM for copy, slides (python-pptx) and website scaffolding (GEN-005, 013, 027).
HW3 (multi-GPU 80 GB+). HW2 plus open video diffusion (Wan, LTX, HunyuanVideo class) for GEN-002, talking-head/avatar and voice-cloning models for GEN-003, virtual try-on diffusion for GEN-010, image-to-3D and text-to-3D generators for GEN-008, batch rendering for GEN-015.
HW4 (cluster). Large video and 3D pipelines integrated with compositing and virtual production (GEN-021): long-clip video models, upscaling and interpolation farms, per-project fine-tuning with consented likeness data.
Cross-family reuse: the same diffusion server serves VIS-027 and LANG-005/014 image needs; TTS, ASR and music services are shared with SPEECH and CONV-018; the vLLM text endpoint and slide/doc parsing are shared with LANG and DOC; GEN-020 outputs feed VIS inspection training; C2PA/watermark and likeness-consent services are shared with ANOM-015 and SEC-011.