Pith. sign in

REVIEW 2 cited by

Diffusion Soup: Model Merging for Text-to-Image Diffusion Models

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2406.08431 v1 pith:NPLDDAPJ submitted 2024-06-12 cs.CV cs.AIcs.CRcs.LG

classification cs.CVcs.AIcs.CRcs.LG
keywords diffusiondatasoupmodelsshardsanti-memorizationdomainenables
verification ladder T0 review T1 audit T2 compute T3 formal

Signed reviews

No signed human review yet.

0 comments
abstract

We present Diffusion Soup, a compartmentalization method for Text-to-Image Generation that averages the weights of diffusion models trained on sharded data. By construction, our approach enables training-free continual learning and unlearning with no additional memory or inference costs, since models corresponding to data shards can be added or removed by re-averaging. We show that Diffusion Soup samples from a point in weight space that approximates the geometric mean of the distributions of constituent datasets, which offers anti-memorization guarantees and enables zero-shot style mixing. Empirically, Diffusion Soup outperforms a paragon model trained on the union of all data shards and achieves a 30% improvement in Image Reward (.34 $\to$ .44) on domain sharded data, and a 59% improvement in IR (.37 $\to$ .59) on aesthetic data. In both cases, souping also prevails in TIFA score (respectively, 85.5 $\to$ 86.5 and 85.6 $\to$ 86.8). We demonstrate robust unlearning -- removing any individual domain shard only lowers performance by 1% in IR (.45 $\to$ .44) -- and validate our theoretical insights on anti-memorization using real data. Finally, we showcase Diffusion Soup's ability to blend the distinct styles of models finetuned on different shards, resulting in the zero-shot generation of hybrid styles.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. DMM: Building a Versatile Image Generation Model via Distillation-Based Model Merging

    cs.CV 2025-04 conditional novelty 6.0 of 10

    DMM trains a single style-promptable diffusion model to reproduce the outputs of multiple teacher models, achieving a merged-model FIDt of 77.51 versus a reference of 74.91.

  2. Decentralized Diffusion Models

    cs.CV 2025-01 conditional novelty 5.0 of 10

    An ensemble of expert diffusion models trained in isolation on disjoint data clusters, combined by a learned router, matches the global flow-matching objective and outperforms a monolithic model at equal FLOPs.

Pith tools