SphereEncoder2 hits SOTA pixel-space image quality with single forward pass
tomgoldsteincs · x · 2026-11-06
Tom Goldstein's team released SphereEncoder2, their 2nd-gen ImageNet model achieving SOTA image quality (FDr6 score) among all known pixel-space image generators, including diffusion models. The Sphere Encoder is a faster diffusion alternative that generates images in a single forward pass — essentially an autoencoder on steroids mapping 100k-dimensional random latent vectors directly to natural images. Sample images trained only on ImageNet-1K show impressive quality.
Related event: SphereEncoder2 Achieves Pixel-Space SOTA in Single Forward Pass(2 posts)→
More from Multimodal
- Maestro v4 ships with built-in MCP server letting ChatGPT, Claude, Codex control it locally or remotely — cocktailpeanut · 2026-11-06
- Local AI video studio Maestro: reference-video workflows with LTX, Wan, Flux and more — cocktailpeanut · 2026-11-06
- Early tests suggest Qwen Image Edit 2.1 underperforms 2511 and 2512 — cal_01 · 2026-11-06
- VESTIGE: a one-day I2V experiment faking a continuous 42-second camera move — magratheya_64 · 2026-11-06
- Recreating GTA 6's Rockstar art style with Magnific + Nano Banana 2.1, prompt included — azed_ai · 2026-11-06
- No image or video models: developer generates visuals, music and SFX purely with Claude-written code — azed_ai · 2026-11-06