PRA hits new image-generation SOTA with 511M parameters and FID 1.94
jiqizhixin · x · 2026-07-21
Researchers from Peking University and DP Technology propose Parallel Rollout Approximation (PRA), a new way to generate images pixel by pixel without a separate tokenizer. PRA first produces compact intermediate states and then decodes them into pixels, which reduces rollout errors while keeping training fast.
- PRA-S (135M) beats the prior billion-scale pixel-space AR model on FID, 2.58 vs 3.60.
- PRA-L (511M) reaches a new state of the art with FID 1.94.
- It also improves classification accuracy over other AR and diffusion models, suggesting a possible unified path for image generation and understanding.
More from Research
- ARISE study tested 45 AI clinical tools in 1,100 consult cases — HealthcareAIGuy · 2026-07-21
- Async OPD distillation doubles throughput while matching synchronous math accuracy — _lewtun · 2026-07-21
- A forecasting lesson on why R-squared alone led to overfitting and worse predictions — mdancho84 · 2026-07-21
- Google DeepMind’s Project Genie talk shows how creatives feed into model research — alexanderchen · 2026-07-21
- Nat Lambert says RL distillation does not use the strongest models as teachers — natolambert · 2026-07-21
- Thread claims GPT-5.6 Sol helped build a new counterexample factory for the Jacobian conjecture — LucaAmb · 2026-07-21