Image generation used Transformers for years; autoregression plus language-only pretraining is the novelty
benno_krojer · x · 2026-09-18
In a technical clarification, bennokrojer points out that Transformers are an architecture and image generation has relied on them for years (versus CNNs/U-Nets). The truly unusual ingredients are autoregression as the generation objective and language-only pretraining — their combination defines the current image generation approach.
More from Models
- Codex Reset Coupon Changes Usage Reset Cycle, User Flags Billing Quirk — algo_diver · 2026-09-19
- 28 hashes logged: crowd audits Gensyn's open-1b, first verifiable training run — benfielding · 2026-09-19
- Open-source auto-router prices every model per request, and shows warm cache beats cheaper models — airesearch12 · 2026-09-19
- Dev builds ErgoPulse, an open-source webcam posture app powered by DeepSeek V4.1 Flash — Arindam_1729 · 2026-09-19
- Bolt adds DeepSeek V4.1 Flash after week one: usage 10x DeepSeek V4 Pro — rohanpaul_ai · 2026-09-19
- Apple researchers propose probe guidance, cutting guidance cost for diffusion LMs with no extra forward pass — itsbautistam · 2026-09-19