Image.

Image is the kernel image brain. It runs four Mickai-native primitives in concert. Atelier handles sovereign text-to-image and image-to-image generation at quality matching the closed frontier models (12B-30B parameter diffusion and flow architectures). Etcher handles inpainting, outpainting, mask-aware editing, and seamless composition. Iris handles vision-language understanding: captioning, classification, OCR, visual question answering. Plinth handles 3D asset synthesis for game, engineering, and pre-visualisation work. Every output carries an ML-DSA-65 signature (patent 08), full lineage in the audit ledger (patent 16), C2PA-grade provenance metadata, and a dual-layer watermark (patent 11): a robust steganographic signal in the pixels plus a cryptographic seal in the metadata. Generation runs entirely on operator-controlled silicon. No image is ever transmitted to a vendor. The structural advantage over closed frontier models is not just sovereign operation but proof: every Mickai image is regulator-verifiable, every closed-vendor image is not.

• 01Sovereign text-to-image, image-to-image, and reference-conditioned generation
• 02Inpainting, outpainting, mask-aware editing, multi-shot character consistency
• 03Vision-language understanding (caption, classify, OCR, visual question answering)
• 043D asset synthesis for game, engineering, and pre-visualisation pipelines
• 05Signed provenance, C2PA-grade metadata, dual-layer watermarking
• 06Liveness-attested avatar still rendering for video coordination (patent 09)

Authoritative external corpora and standards this brain treats as canonical. Every retrieval against these sources is signed into the audit ledger so a regulator can prove which evidence drove which output.

• 01Mickai Patents 08, 09, 11, 16
• 02Flow-matching and rectified-flow diffusion architecture research
• 03Latent-diffusion model research corpus
• 04Frontier text-to-image architecture research
• 05C2PA Content Provenance and Authenticity specification
• 06Steganographic and post-hoc image-watermarking research
• 073D asset synthesis research (mesh, gaussian-splat, point-cloud)
• 08Diffusion-model academic corpus (DDPM, DDIM, Score Matching, Flow Matching)
• 09Vision-language model research (open-weights frontier)
• 10ITU-T H.273 colour parameter spec and ICC colour profile spec
• 11EXIF metadata standard, PNG and JPEG image-format specifications

Mickai-native tooling primitives this brain implements internally. Codex for sovereign plain-text graph PKM, Lectern for spaced-repetition memory, Stele for citation-provenance, and domain-native primitives layered on top. No external services in the trust path; data stays on operator-personalised hardware.

• 01Atelier (sovereign image generation engine)
• 02Etcher (inpainting, outpainting, mask-aware editing)
• 03Iris (vision-language understanding)
• 04Plinth (3D asset synthesis)
• 05Marble (composition canvas)
• 06Cataloguer (on-device image management)
• 07Wax-Seal (signed image provenance and watermarking)