AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Improving Guidance-Free Visual Generation via Self-Contrastive Alignment for Likelihood Estimation

Jiwan Hur, DongJae Lee, Dongyeun Lee, Gyojin Han, Junmo Kim

NeurIPS 2026 · Image generation › Fast diffusion sampling · approach: Distributional control of diffusion

Read the paper

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes self-contrastive guidance (SCG) for guidance-free visual generation, contrasting the model's own predictions to approximate CFG at single-inference cost.

Problem

Classifier-free guidance improves visual generation quality but doubles sampling cost; existing guidance-free methods that approximate CFG with one inference are bounded by CFG's performance and inherit its limitations.

Approach

Self-contrastive guidance contrasts the model's prediction on its own inputs (the abstract is truncated, so the exact contrast is not stated) to get guidance without a separate unconditional pass, aiming to exceed the CFG-approximation ceiling.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Add self-contrastive guidance (SCALE) on top of GFT for diffusionDiT-XL/2 guidance-free · ImageNet 256x256FID-50k: 1.99 -> 1.87Improved
Add SCALE on top of GFT for autoregressive modelLlamaGen-L guidance-free · ImageNet 256x256FID-50k: 2.88 -> 2.58Improved
Use SCALE instead of CCA on LlamaGen-LLlamaGen-L · ImageNet 256x256IS: 288.2 -> 253.0Hurt
Replace two-pass CFG with guidance-free SCALEDiT-XL/2 · ImageNet 256x256FID: 2.17 -> 1.87Improved
Lower alpha from 0.8 to 0.25 at sampling (ablation)SCALE · SD 1.5 text-to-imageFID-30k: 12.83 -> 16.52Hurt
Lower alpha from 0.8 to 0.25 at sampling (ablation)SCALE · SD 1.5 text-to-imageHPSv2.1: 0.253 -> 0.261Improved
Apply DPO to guidance-free base SD 1.5SD 1.5 guidance-free · SD 1.5 text-to-imagepreference metricsNo clear effect
Add DPO on top of SCALESCALE · SD 1.5 text-to-imagepreference metricsImproved
Train SCALE from scratch instead of GFTDiT-B/2 · ImageNet 256x256 (from scratch)FID: 9.04 -> 8.75Improved
Use patch-wise mix instead of linear interpolation (ablation)SCALE · DiT-B/2 ImageNetFID: 8.67 -> 8.60Improved
Lower PatchMix ratio lambda from 1.0 to 0.2 (ablation)SCALE · LlamaGen-L ImageNetIS: 245.9 -> 253.0Improved
Lower PatchMix ratio lambda from 1.0 to 0.2 (ablation)SCALE · LlamaGen-L ImageNetFID: 2.59 -> 2.58No clear effect

Limitations

Experimental setup

Baselines: classifier-free guidance guidance-free approaches that approximate CFG

Builds on: classifier-free guidance

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy