Personal Image Model Studio 0.2.0 — private local CPU app

Personal Image Model Studio 0.2.0 — Jason Stys

Real local accounts and a portable native CPU diffusion application. Open the
executable, register, choose an independent vault phrase, create owned synthetic
data, train actual weights and generate learned images—no Python/npm setup for
that native workflow. No third-party/pretrained weights or GPU drivers are bundled.

• Account-scoped records/images, scrypt login hashes, HttpOnly sessions, recovery,
  explicit deletion, origin/CSRF/body limits and private per-OS runtime storage.
• AES-256-GCM locked archives; the key exists in the trusted local service while
  unlocked. ML uses temporary plaintext files. Crashes can leave owner-bound
  recovery data. Use device encryption and a private OS login.
• File-picker imports with ordered 4-MiB chunks; arbitrary PC paths, browser-selected
  executables, pickle and executable models are blocked in account mode.
• Preserved native training/resume, generation, exact-known-pixel masked edits,
  feedback/explicit revisions and optional consented own-gallery connector.
• Built-in help, rights/privacy/install limits and Jason Stys's public contacts.
• Checksummed Windows amd64, macOS arm64 and Linux x86_64 CPU archives, published
  only after exact-source Validate passes and all three binaries pass genuine
  optimizer/sampler/account/encrypted-reload smoke tests.

Limits: unsigned portable executables, not signed installers/notarized apps.
One unlocked account/local engine; 8 GiB/20,000 workspace files. The compact model
is not a general/photorealistic generator. Built-in CPU mode supports native models;
classic SD/LoRA remains available with an explicitly trusted full source/runtime.
An encrypted archive alone does not restore login/key settings: cold-back up the
entire protected account directory. Forgot the vault phrase? There is no recovery
backdoor. PNG/report/feedback exports are plaintext. No public host or native mobile
package is provisioned. Browser UI is responsive; OS/device compatibility is not
universal. Existing developer data is not silently migrated.

See installation,
architecture decision
and validation. Contact
Jason Stys; never put private art,
credentials, keys or provider tokens into a public issue.