Project ideas from Hacker News discussions.

What an unprocessed photo looks like

๐Ÿ“ Discussion Summary (Click to expand)

1. Extensive Camera Processing is Essential

All photos require processing like debayering, gamma correction, and tone mapping to produce viewable images from raw sensor data.
">Thereโ€™s nothing that happens when you adjust the contrast or white balance in editing software that the camera hasnโ€™t done under the hood." - stavros
"Everything is an interpretation of the data that the camera has to do, making a thousand choices along the way." - stavros

2. No "Real" vs. "Fake" Photos; Intent Matters

Photos are always processed interpretations; "fakeness" depends on deceptive intent, not edits like global filters vs. local AI changes.
"Fake images are images with intent to deceive. By that definition, even an image that came straight out of the camera can be 'fake'." - stavros
"Intent is the defining factor... If you dial down the exposure to make the photo more dramatic, you're manipulating emotions too." - nospice

3. Bayer Filter and Human Perception Drive Design

Sensors use RGGB patterns (50% green) for luminance detail, matching eye sensitivity; alternatives like Foveon exist but are niche.
"The reason the Bayer pattern is RGGB (50% green) isn't just about color balance, but spatial resolution. The human eye is most sensitive to green light." - barishnamazov
"Green spectral response curve is close to the luminance curve humans see... Twice the pixels to increase the effective resolution in the green/luminance channel." - milleramp


๐Ÿš€ Project Ideas

RAW Processing Playground

Summary

  • Interactive web app where users upload RAW files and apply processing steps (debayering, white balance, gamma, tone mapping) via sliders, visualizing each stage side-by-side with the camera's output.
  • Core value proposition: Demystifies the "black box" of camera pipelines, letting users experiment to create custom "honest" renditions without painted looks or over-processing.

Details

Key Value
Target Audience Amateur photographers, HN tech enthusiasts debugging sensor data
Core Feature Step-by-step pipeline with real-time previews, export to DNG/JPEG
Tech Stack WebAssembly (libraw.js for RAW decode), Three.js/WebGL for visualization, React for UI
Difficulty Medium
Monetization Revenue-ready: Freemium (basic free, pro exports $5/mo)

Notes

  • Addresses throw310822's frustration with poor examples: "pity the author chose such a poor example... making it really hard to understand what the 'ground truth'". HN loves interactive tools for grokking complex pipelines.
  • High discussion potential as users share custom profiles; practical for astrophotographers or RAW tinkerers.

SensorTune ML ISP

Summary

  • Desktop app using ML models trained per-camera model for joint debayering/denoising/HDR stacking on RAW bursts, with adjustable "honesty" sliders (e.g., noise preservation vs smoothing).
  • Core value proposition: Brings phone-like computational photography to DSLRs/full-frame RAWs, unlocking low-ISO quality from high-ISO bursts without "painted" artifacts.

Details

Key Value
Target Audience DSLR/mirrorless owners, computational photography hobbyists
Core Feature Burst stacking with motion compensation, sensor-specific models (trainable via dark/flat frames)
Tech Stack PyTorch/TensorFlow for models, libraw/darktable backend, Electron for GUI
Difficulty High
Monetization Revenue-ready: One-time $29 license

Notes

  • Solves jiggawatts' pain: "very little of the computation photography magic... applied to larger DSLRs"; quotes krackers on bit depth/linear light. HN would love open models for community training.
  • Practical utility for low-light shooters; sparks threads on sensor calibration.

HonestCam Toggle Suite

Summary

  • Mobile app/plugin for phone cameras exposing RAW with user-toggleable pipeline stages (disable AI denoising/skin smoothing/object removal), plus medical mode for accurate skin/lesion capture.
  • Core value proposition: Gives control over "fake" processing, ensuring correlation to sensor data for journalism, medical use, or purists wanting "honest noise".

Details

Key Value
Target Audience Phone photographers frustrated by over-processing, medical users
Core Feature Real-time RAW preview with toggles (e.g., no face enhancement), batch export
Tech Stack Camera2 API (Android)/AVFoundation (iOS), ONNX for lightweight ML toggles
Difficulty Medium
Monetization Hobby

Notes

  • Tackles NiloCK's rash issue and trinix912's "borderline generative AI": "photos were always 'nicer' than what my eyes recorded". Aligns with card_zero's "fake cosmetic bullshit" aversion.
  • Utility for evidence/journalism; HN debates "real vs fake" would drive adoption/sharing.

Read Later