Updates the reference image to the latest version. Removes the barefoot
instruction from the character prompt and clarifies that vampire fangs
are only visible when the mouth is open.
Replaces the boring timestamp/prompt-truncation thread naming with randomly generated adjective-noun combos (e.g. *Crimson Reverie*, *Neon Phantom*, *Gilded Silence*).
50 adjectives × 50 nouns = 2,500 unique possible thread names, applied consistently across all three modes.
✨ This issue was created with help from Hikari~ 🌸
Reviewed-on: #19
Co-authored-by: Hikari <hikari@nhcarrigan.com>
Co-committed-by: Hikari <hikari@nhcarrigan.com>
## Summary
- Adds a two-step thread creation modal — step 1 picks the mode, step 2 configures generation options
- Art mode now supports user-selectable aspect ratio (1:1, 4:3, 3:4, 16:9, 9:16, 21:9)
- All three modes (Art, Avatar, Replace) now support user-selectable resolution (1K, 2K, 4K)
- Mode label in the input area reflects the chosen settings (e.g. `🩷 Art Mode (16:9) · 4K`)
- Backend cost calculation now scales with resolution
- Regenerated `icon.ico` with clean BMP-only entries to fix Windows local builds
✨ This PR was created with help from Hikari~ 🌸
Reviewed-on: #18
Co-authored-by: Hikari <hikari@nhcarrigan.com>
Co-committed-by: Hikari <hikari@nhcarrigan.com>
## Summary
The Windows build job has been failing on every PR since the initial release. The root cause is that the default `clang`/`lld`/`llvm` apt packages resolve to an older version of `llvm-rc` that cannot handle PNG-compressed entries in `.ico` files — and all six images in `icon.ico` use PNG compression.
Pinning to `clang-18`, `lld-18`, and `llvm-18` (available in Ubuntu 24.04's default repos) and registering them via `update-alternatives` ensures `llvm-rc`, `clang-cl`, `lld-link`, and friends all resolve to a version that handles the icon resource correctly.
✨ This PR was created with help from Hikari~ 🌸
Reviewed-on: #17
Co-authored-by: Hikari <hikari@nhcarrigan.com>
Co-committed-by: Hikari <hikari@nhcarrigan.com>
## Summary
- Art and Avatar modes now support optional image uploads (paste or file picker), sent to Gemini alongside the text prompt
- Initial Replace mode gains an optional textarea for notes to include with the uploaded image
- Backend updated to send user images for all modes, appending replace instructions only when in replace mode
- Send validation updated to allow image-only messages in art and avatar modes
✨ This PR was created with help from Hikari~ 🌸
Reviewed-on: #16
Co-authored-by: Hikari <hikari@nhcarrigan.com>
Co-committed-by: Hikari <hikari@nhcarrigan.com>
Tatsumi is a Tauri desktop app for generating AI character art of Naomi
using Google Gemini's image model. Features three generation modes
(avatar, art, replace), persistent conversation threads, message
editing and deletion, retry support, cost tracking, and an about modal
with lore-accurate self-introduction from Emi Carrigan.
hikari
naomi
NHCarrigan