Oriel — AI & technology worth noticing, every day.

Intrinsic Distribution Matching

Paper · https://arxiv.org/abs/2610.39633

One-step and few-step text-to-image generation must accurately render the details specified in the prompt despite limited opportunities for refinement. Distribution matching distillation (DMD) uses score differences to bring the generated distribution closer to a reference distribution. Following prior work, we estimate this difference with internal velocity branches, avoiding a separate fake-score network. We ask whether an explicit moment constraint improves this learned score supervision. A decomposition of the reverse-KL field into a Gaussian moment component and a non-Gaussian residual identifies statistics that can be constrained explicitly. We introduce Intrinsic Distribution Matching (IDM), which retains the full DMD correction and matches feature means and per-coordinate standard deviations using the generator's exponential-moving-average (EMA) copy, requiring no external feature encoder. Experiments use full-parameter and low-rank adaptation (LoRA) tuning on Z-Image-Turbo 6B and Qwen-Image 20B. At two network evaluations, IDM reaches 94.25% and 90.20% on Chinese and English LongText-Bench, the highest scores among the compared methods at this budget. Controlled ablations show gains from the added moment loss, while generator EMA features perform competitively with an external DINOv2 encoder across the five reported metrics. Code is available at https://github.com/inclusionAI/IDM.

First seen by Oriel on .

AI summary

Intrinsic Distribution Matching (IDM) augments distribution matching distillation with explicit feature-moment constraints, improving text-to-image generation quality at two network evaluations.

  • IDM retains the full DMD correction and matches feature means and per-coordinate standard deviations using the generator’s EMA copy, without an external feature encoder.
  • The method decomposes the reverse-KL field into a Gaussian moment component and a non-Gaussian residual to identify statistics for explicit constraints.
  • Experiments cover full-parameter and LoRA tuning on Z-Image-Turbo 6B and Qwen-Image 20B.
  • At two network evaluations, IDM scores 94.25% on Chinese LongText-Bench and 90.20% on English LongText-Bench, the highest among compared methods at that budget.
  • Ablations attribute gains to the added moment loss, while EMA features perform competitively with an external DINOv2 encoder across five reported metrics.

#text-to-image generation #distribution matching #model distillation #low-rank adaptation

Generated by gpt-5.6-sol. Check the original for details.

Metric trends

Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained

Points

Score reported by the source.

  • Hugging Face Papers: 30, latest observation ; -204 over observed period since

Comments

Comment count reported by the source.

  • Hugging Face Papers: 4, latest observation ; +2 over observed period since

Aliases

SourceExternal IDURL
Hugging Face Papers2610.39633https://arxiv.org/abs/2610.39633 Read discussion