{"paper":{"title":"SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformers","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"Sana-0.6B generates high-resolution images competitively with 12B-parameter models while running over 100 times faster on consumer GPUs.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Enze Xie, Han Cai, Haotian Tang, Junsong Chen, Junyu Chen, Ligeng Zhu, Muyang Li, Song Han, Yao Lu, Yujun Lin, Zhekai Zhang","submitted_at":"2024-10-14T15:36:42Z","abstract_excerpt":"We introduce Sana, a text-to-image framework that can efficiently generate images up to 4096$\\times$4096 resolution. Sana can synthesize high-resolution, high-quality images with strong text-image alignment at a remarkably fast speed, deployable on laptop GPU. Core designs include: (1) Deep compression autoencoder: unlike traditional AEs, which compress images only 8$\\times$, we trained an AE that can compress images 32$\\times$, effectively reducing the number of latent tokens. (2) Linear DiT: we replace all vanilla attention in DiT with linear attention, which is more efficient at high resolu"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Sana-0.6B is very competitive with modern giant diffusion model (e.g. Flux-12B), being 20 times smaller and 100+ times faster in measured throughput.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The 32x deep-compression autoencoder preserves sufficient perceptual quality and text alignment at high resolutions without introducing artifacts that linear attention cannot correct.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Sana-0.6B produces high-resolution images with strong text alignment at 20x smaller size and 100x higher throughput than Flux-12B by combining 32x image compression, linear DiT blocks, and a decoder-only LLM text encoder.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Sana-0.6B generates high-resolution images competitively with 12B-parameter models while running over 100 times faster on consumer GPUs.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b59caa39f172598343a38d0cc619ff6837b0e3aaf67b4883cf3057076ac0fbaa"},"source":{"id":"2410.10629","kind":"arxiv","version":3},"verdict":{"id":"dc9f5c27-201c-4f60-b8ef-12c3f83aef97","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T00:52:14.386983Z","strongest_claim":"Sana-0.6B is very competitive with modern giant diffusion model (e.g. Flux-12B), being 20 times smaller and 100+ times faster in measured throughput.","one_line_summary":"Sana-0.6B produces high-resolution images with strong text alignment at 20x smaller size and 100x higher throughput than Flux-12B by combining 32x image compression, linear DiT blocks, and a decoder-only LLM text encoder.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The 32x deep-compression autoencoder preserves sufficient perceptual quality and text alignment at high resolutions without introducing artifacts that linear attention cannot correct.","pith_extraction_headline":"Sana-0.6B generates high-resolution images competitively with 12B-parameter models while running over 100 times faster on consumer GPUs."},"references":{"count":29,"sample":[{"doi":"","year":null,"title":"eDiff-I: Text-to-Image Diffusion Models with an Ensemble of Expert Denoisers","work_id":"2cd7b629-ab37-4ce5-b51e-aa4d99547468","ref_index":1,"cited_arxiv_id":"2211.01324","is_internal_anchor":true},{"doi":"","year":2022,"title":"All are worth words: a vit backbone for score-based diffusion models","work_id":"98f35013-1df4-4a37-8b71-6206599f9833","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2005,"title":"Language Models are Few-Shot Learners","work_id":"214732c0-2edd-44a0-af9e-28184a2b8279","ref_index":3,"cited_arxiv_id":"2005.14165","is_internal_anchor":true},{"doi":"","year":2020,"title":"Mahoney, and Kurt Keutzer","work_id":"911a6b35-56a0-42b3-9951-b6f6ac9da2ed","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"PixArt- : Weak-to-strong training of diffusion transformer for 4k text-to-image generation","work_id":"5dc0cfb7-2014-4631-be70-f632bf6ca9ec","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":29,"snapshot_sha256":"8dab2fbac73eaa14a78aa7dcdf69c72b55b8576ffa2200f5345add58a55a44ee","internal_anchors":11},"formal_canon":{"evidence_count":1,"snapshot_sha256":"a7925ba5afbca173b270de0c9e79d4db4aa675988ba6f9d8760e372d6fde38d2"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}