Blending Concepts in Text-to-Image Diffusion Models using the Black Scholes Algorithm

ICLR 2025 Conference Submission1678 Authors

Keywords: content creation, diffusion models

TL;DR: An approach for blending concepts in diffusion models using novel perspectives from the Black Scholes model in economics and finance.

Abstract: Many image generation tasks, such as content creation, editing, personalization, and zero-shot generation, require generating unseen concepts without retraining the model or collecting additional data. These tasks often involve blending existing concepts by conditioning the diffusion model with text prompts at each denoising step, a process known as ``prompt mixing''. We introduce a novel approach for prompt mixing to forecasts predictions w.r.t. the generated image and makes informed text conditioning decisions at each time step during diffusion denoising. To do so, we leverage the connection between diffusion models (rooted in non-equilibrium thermodynamics) and the Black-Scholes model for pricing options in Finance, and draw analogies between the variables in both contexts to derive an appropriate algorithm for prompt mixing using the Black Scholes model. Specifically, the parallels between diffusion models and the Black-Scholes model enable us to leverage properties related to the dynamics of the Markovian model derived in the Black-Scholes algorithm. Our prompt-mixing algorithm is data-efficient, meaning it does not need additional training. Furthermore, it operates without human intervention or hyperparameter tuning. We highlight the benefits of our approach by comparing it, qualitatively and quantitatively using CLIP scores, to other prompt mixing techniques, including linear interpolation, alternating prompts, step-wise prompt switching, and CLIP-guided prompt selection across various scenarios such as single object per text prompt, multiple objects per text prompt and objects against backgrounds. The resulting code will be made publicly available for research reproduction.

Primary Area: generative models

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 1678