Show HN: Wall-mounted diffusion mirror that turns reflections into paintings
A wall-mounted “diffusion mirror” that uses a camera and generative image models to turn real-time reflections into painterly images is drawing interest both as a technical project and as an artwork. Commenters weigh in on practical aspects such as frame rate, local vs cloud processing, energy use, privacy, and hardware choices, while suggesting enhancements like interpolation, freezing favorite images, networking multiple frames, or using e-ink displays. The project also prompts a wider debate about what counts as art, how much artistic value lies in the automated images versus the installation itself, and whether AI tools democratize expression or bypass the transformative effort of learning a craft.
Overall reaction
- Many commenters find the diffusion “mirror” delightful and would like one at home or in public spaces.
- Some see the low frame rate and instability as part of its charm; others view them as technical shortcomings.
Frame rate, visual style, and interaction
- Debate over framerate:
- Some argue for higher FPS via morphing, crossfades, interpolation, or batching to reduce “popping.”
- Others say low update rates (seconds to minutes) suit art better, giving time to appreciate each image and saving energy.
- Suggestions include:
- Transitional effects between images.
- Freezing or saving particularly good frames (e.g., for avatars or lockscreens).
- Using it more like a slow “capture once, display until next shot” device, which would no longer be a literal mirror.
Architecture, performance, hardware, and privacy
- Technical tips:
- Use binary WebSocket payloads instead of base64; possibly send raw RGB instead of JPEG.
- Cache text embeddings; consider img‑to‑img fine‑tuning and models like SDXL Turbo/Lightning.
- Queue frames and batch processing as in StreamDiffusion.
- Hardware discussions:
- Raspberry Pi vs Jetson Nano vs running locally on GPUs.
- Alternatives like e‑ink displays for low power, and using old tablets as cheaper displays.
- Strong concern about streaming webcam feeds to third‑party cloud GPUs; several want everything local, ideally in-frame.
- Some wonder about hardware accelerators (e.g., Coral), but note incompatibilities with common image models.
- Infrared camera/light is used for working in the dark and adds a wand‑like interaction mode.
Art, skill, and authorship
- Some claim the installation as a whole is the true artwork; the individual AI frames are “output,” not art.
- Disagreement over whether art is mainly:
- “Surfacing the inner world” vs.
- Intentional, skilled craft vs.
- Evoking or preserving emotion vs.
- Practice and process over tools.
- Several argue that AI lowers barriers for people with ideas but limited skill; others insist the skill‑building journey is essential and irreplaceable.
- Some view this device as “dynamic/immersive art” or “wallpaper,” not deep psychological expression.
Extensions and alternative concepts
- Ideas floated:
- Virtual camera feed for video calls.
- Sentiment/toxicity‑driven transformations (e.g., Dorian Gray effect).
- Networking multiple frames so viewers see strangers through the style transform (“art chat roulette”).
- Moving the camera elsewhere, or into other homes, to decouple “mirror” from self‑reflection.
- Occasional playful or eerie inpainting (e.g., figures behind the viewer).