Diffusion Community Revisits End-to-End Latent Diffusion Training in REPA Lineage
serrjoa · x · 2026-11-01
A discussion thread tracing the diffusion training paradigm since REPA, which aligned denoiser hidden states with external visual representations and sped up SiT training by over 17.5x. REPA-E extended this with end-to-end DiT+VAE training, with the key finding that backpropagating diffusion loss into the VAE causes latent mode collapse. The reposters argue that despite ImageNet DiT being a toy setting now, the end-to-end training approach remains worth studying for its simplicity and elegance.
More from Research
- Open-source memory layer ICE loses to vector-RAG on LongMemEval: 43% vs 69.5% — Deepnar_S · 2026-11-01
- Tiny 109M model beats 72B models at image-text alignment with 42x less data — Lazylion2 · 2026-11-01
- Can you name these three ML architectures from blank diagrams? — cneuralnetwork · 2026-11-01
- TIL: Boolean circuits can be cyclic — NOR-gate latches and Rivest-style loops — thomasahle · 2026-11-01
- Platypus: an open-source continual-learning LM that infers new facts while it "sleeps" — ChrisLLM · 2026-11-01
- The Path of Averaging: Tracing the Line from Attention to TTT — SonglinYang4 · 2026-11-01