AI Image Generation on a 1980s Tandy 286: Retro Rig Renders Pics in 9 Seconds
jacobpederson · reddit · 2026-10-01
A developer got AI image generation running on a 10 MHz Tandy 1000 TL/3 (286): a DOS program called DeskMind talks over PicoMEM WiFi to a Python server that calls Krea 2 in ComfyUI, with Qwen3.8-27B rewriting prompts — about 9 seconds per image.
Key techniques:
- Dither-aware prompt rewriting: Qwen rewrites prompts to request one large subject, bold shapes, strong contrast, and no fine detail, since the 16-color RGBI palette punishes gradients.
- Anamorphic resize before dithering: 640x200 pixels are 2.4x taller than wide on a 4:3 CRT, so images are squashed with Lanczos first or circles come out wrong.
- Pre-dither grading: contrast 1.15, saturation 1.3, light unsharp mask at target size.
- Three dither engines (Floyd-Steinberg, Atkinson, Bayer, Yliluoma, etc.) with a live preview GUI.
- Vision sees the dither too: the original and dithered images are both fed to Qwen in chat.
Output is a 160x50 thumbnail plus 64,000 bytes of packed 4-bit pixels written straight to video memory (1s transfer). GPLv3: github.com/RowanUnderwood/DeskMind
Related event: Developer Runs AI Chat and Image Generation on a 40-Year-Old 286 PC(2 posts)→
More from coding & agent
- shadcn boosts @uisdk: one-shot UI rebuild that's instantly E2E-testable by AI — shadcn · 2026-11-16
- AI agent deletes 43,599 lines of OpenClaw test code in one PR — vincent_koc · 2026-11-16
- JPMorgan shows Jarvis, its internal agent assembly line, at Interrupt NYC — LangChain · 2026-11-16
- Dev clarifies: Apple doesn't ban feature flags, only fooling App Review — talkaboutdesign · 2026-11-16
- Antigravity Matches Claude Cowork in Tests, While Gemini Tasks' Purpose Remains Unclear — sebajun2 · 2026-11-16
- One 16GB GPU runs songs, voice, images, 720p video and local RAG — all offline and open source — llama-of-death · 2026-11-16