# Direct Ascent Synthesis: Revealing Hidden Generative Capabilities in Discriminative Models

arXiv: https://arxiv.org/abs/2502.07753 · HTML: https://mlsanity.com/paper/2502.07753/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 11 Feb 2025.

## Built on by (most cited first, 8 of 8)

- [When Pretty Isn't Useful: Investigating Why Modern Text-to-Image Models Fail as Reliable Training Data Generators](https://mlsanity.com/paper/2602.19946/) (2026)
- [Implicit Inversion turns CLIP into a Decoder](https://mlsanity.com/paper/2505.23161/) (2025)
- [Representations of Text and Images Align From Layer One](https://mlsanity.com/paper/2601.08017/) (2026)
- [PRISM: Diversifying Dataset Distillation by Decoupling Architectural Priors](https://mlsanity.com/paper/2511.09905/) (2025)
- [TRANSPORTER: Transferring Visual Semantics from VLM Manifolds](https://mlsanity.com/paper/2511.18359/) (2025)
- [Efficient Dataset Distillation for Pre-Trained Self-Supervised Models via Statistical Flow Matching](https://mlsanity.com/paper/2602.05391/) (2026)
- [QoS-Aware Token Scheduling and Private Data Valuation for Multi-Modal Agentic Networks](https://mlsanity.com/paper/2606.15573/) (2026)
- [Anchoring and Steering Diffusion: Enhancing the Faithfulness of Text-to-Image Generation at Inference Time](https://mlsanity.com/paper/2607.26647/) (2026)

## Abstract (arXiv)

We demonstrate that discriminative models inherently contain powerful generative capabilities, challenging the fundamental distinction between discriminative and generative architectures. Our method, Direct Ascent Synthesis (DAS), reveals these latent capabilities through multi-resolution optimization of CLIP model representations. While traditional inversion attempts produce adversarial patterns, DAS achieves high-quality image synthesis by decomposing optimization across multiple spatial scales (1x1 to 224x224), requiring no additional training. This approach not only enables diverse applications -- from text-to-image generation to style transfer -- but maintains natural image statistics (1/f2 spectrum) and guides the generation away from non-robust adversarial patterns. Our results demonstrate that standard discriminative models encode substantially richer generative knowledge than previously recognized, providing new perspectives on model interpretability and the relationship between adversarial examples and natural image synthesis.