Reproducible low-level artifacts found in ChatGPT image generation
DickHorner · reddit · 2026-08-17
A user discovered reproducible, canvas-aligned texture artifacts while testing ChatGPT's image generation and iterative editing features.
Phenomenon:
- Faint cloudy/mottled textures appear in smooth areas (backgrounds, skin) after multiple rounds of generative editing.
- Repeated edits reinforce the texture, while some areas (faces) appear protected/stable.
Key Experiments:
- Image Shifting: Shifting the image by 20px before altering changed the artifact intensity, suggesting it's tied to canvas coordinates, not random noise.
- Black Image Test: Multiple independently generated "black" images showed significant correlation in non-zero pixels (correlation 0.848, Jaccard overlap 0.766), much higher than expected for random independent pixels.
This suggests an underlying spatial pattern or processing mechanism within the model causing non-random artifacts.
More from Multimodal
- Open-source anatomy-reveal skill turns Claude into an exploded-view motion design studio — msharmas · 2026-10-16
- Flux 3 Klein vs Flux 2 Klein hands-on: the new base model 'barely usable' for T2I — Both-Rub5248 · 2026-10-16
- USTC paper uses speculative sampling to cut latency in interactive video generation — ustc · 2026-10-16
- 800,000 Gaussian splats run in WebGPU; Casado sees near-infinite scaling — generativist · 2026-10-16
- Nano Banana 2.1 image generation now live in Perplexity Computer for all users — ChrisUniverse · 2026-10-16
- Demo shows precise camera-path control in 3D scenes from a single input image — XRarchitect · 2026-10-16