DVdvrp·Sep 18, 2023Submission
InstaFlow: ultrafast 1-step Stable Diffusion
3 points · Open discussiongnobitab
InstaFlow
Python2023–2023
gnobitab / InstaFlow
InstaFlow
:zap: InstaFlow! One-Step Stable Diffusion with Rectified Flow (ICLR 2024)
2Distinct people
2Mentions
0Explicit recommenders
1.4KGitHub stars
What to know before exploring
What it is
:zap: InstaFlow! One-Step Stable Diffusion with Rectified Flow (ICLR 2024)
GitHub
Where it is used
No documented product use has been verified yet.
What people are saying
No recommendation phrase was found in the indexed comments.
No classified warning is attached to this page. Read the discussions before treating silence as approval.
Also in these discussions
No other repository is repeatedly linked in the same discussions.
This week on Hacker News
Nov 7–Nov 14 UTC · Direct repository links and separately labeled related project stories. Indexed through 11/14/2026, 3:56:07 PM UTC.
No direct repository links were found in the indexed HN discussions over the past seven days. Older discussions are available below.
Mentions over time
HN mentions by 6-hour interval2 dated mentions
Observed mentions: 2023-09-14 02:48:27 UTC to 2023-09-18 22:47:59 UTC.
Counts come from the repository’s dated source evidence. Index cutoff: 2026-11-02 08:16:59 UTC.
View mention counts and date ranges
| Period | Start (UTC) | End (exclusive, UTC) | Mentions | Submissions | Comments |
|---|---|---|---|---|---|
| Sep 14, 00:00 | 2023-09-14 00:00:00 UTC | 2023-09-14 06:00:00 UTC | 1 | 0 | 1 |
| Sep 14, 06:00 | 2023-09-14 06:00:00 UTC | 2023-09-14 12:00:00 UTC | 0 | 0 | 0 |
| Sep 14, 12:00 | 2023-09-14 12:00:00 UTC | 2023-09-14 18:00:00 UTC | 0 | 0 | 0 |
| Sep 14, 18:00 | 2023-09-14 18:00:00 UTC | 2023-09-15 00:00:00 UTC | 0 | 0 | 0 |
| Sep 15, 00:00 | 2023-09-15 00:00:00 UTC | 2023-09-15 06:00:00 UTC | 0 | 0 | 0 |
| Sep 15, 06:00 | 2023-09-15 06:00:00 UTC | 2023-09-15 12:00:00 UTC | 0 | 0 | 0 |
| Sep 15, 12:00 | 2023-09-15 12:00:00 UTC | 2023-09-15 18:00:00 UTC | 0 | 0 | 0 |
| Sep 15, 18:00 | 2023-09-15 18:00:00 UTC | 2023-09-16 00:00:00 UTC | 0 | 0 | 0 |
| Sep 16, 00:00 | 2023-09-16 00:00:00 UTC | 2023-09-16 06:00:00 UTC | 0 | 0 | 0 |
| Sep 16, 06:00 | 2023-09-16 06:00:00 UTC | 2023-09-16 12:00:00 UTC | 0 | 0 | 0 |
| Sep 16, 12:00 | 2023-09-16 12:00:00 UTC | 2023-09-16 18:00:00 UTC | 0 | 0 | 0 |
| Sep 16, 18:00 | 2023-09-16 18:00:00 UTC | 2023-09-17 00:00:00 UTC | 0 | 0 | 0 |
| Sep 17, 00:00 | 2023-09-17 00:00:00 UTC | 2023-09-17 06:00:00 UTC | 0 | 0 | 0 |
| Sep 17, 06:00 | 2023-09-17 06:00:00 UTC | 2023-09-17 12:00:00 UTC | 0 | 0 | 0 |
| Sep 17, 12:00 | 2023-09-17 12:00:00 UTC | 2023-09-17 18:00:00 UTC | 0 | 0 | 0 |
| Sep 17, 18:00 | 2023-09-17 18:00:00 UTC | 2023-09-18 00:00:00 UTC | 0 | 0 | 0 |
| Sep 18, 00:00 | 2023-09-18 00:00:00 UTC | 2023-09-18 06:00:00 UTC | 0 | 0 | 0 |
| Sep 18, 06:00 | 2023-09-18 06:00:00 UTC | 2023-09-18 12:00:00 UTC | 0 | 0 | 0 |
| Sep 18, 12:00 | 2023-09-18 12:00:00 UTC | 2023-09-18 18:00:00 UTC | 0 | 0 | 0 |
| Sep 18, 18:00 | 2023-09-18 18:00:00 UTC | 2023-09-19 00:00:00 UTC | 1 | 1 | 0 |
What Hacker News says
2 sourcesBasically Stable Diffusion can be trained to only require one inference step for image generation rather than tens of steps.
Abstract:
Diffusion models have revolutionized text-to-image generation with its exceptional quality and creativity. However, its multi-step sampling process is known to be slow, often requiring tens of inference steps to obtain satisfactory results. Previous attempts to improve its sampling speed and reduce computational costs through distillation have been unsuccessful in achieving a functional one-step model. In this paper, we explore a recent method called Rectified Flow, which, thus far, has only been applied to small datasets. The core of Rectified Flow lies in its \emph{reflow} procedure, which straightens the trajectories of probability flows, refines the coupling between noises and images, and facilitates t
Read the rest
he distillation process with student models. We propose a novel text-conditioned pipeline to turn Stable Diffusion (SD) into an ultra-fast one-step model, in which we find reflow plays a critical role in improving the assignment between noise and images. Leveraging our new pipeline, we create, to the best of our knowledge, the first one-step diffusion-based text-to-image generator with SD-level image quality, achieving an FID (Frechet Inception Distance) of 23.3 on MS COCO 2017-5k, surpassing the previous state-of-the-art technique, progressive distillation, by a significant margin (37.2 → 23.3 in FID). By utilizing an expanded network with 1.7B parameters, we further improve the FID to 22.4. We call our one-step models \emph{InstaFlow}. On MS COCO 2014-30k, InstaFlow yields an FID of 13.1 in just 0.09 second, the best in ≤0.1 second regime, outperforming the recent StyleGAN-T (13.9 in 0.1 second). Notably, the training of InstaFlow only costs 199 A100 GPU days. Project page: https://github.com/gnobitab/InstaFlow