New Workflow Combines AI Tools to Replicate Image Styles
A new workflow for replicating image styles has been developed using a combination of AI tools. The process involves uploading a source image to Nano Banana Pro to generate a style reference (SREF), using Gemini AI to create a detailed prompt, and then iterating to produce images. This technique is already being productized as a service for creating custom portraits.
- The underlying technique is a deep learning method called Neural Style Transfer, which uses convolutional neural networks (CNNs) to mathematically separate the "content" representation of one image from the "style" representation of another, allowing them to be recombined. - This workflow's use of a style reference image (SREF) is similar to Midjourney's `--sref` parameter, which can reference either an image URL or a numerical code corresponding to a specific aesthetic. A key difference is that Midjourney's community actively discovers and shares SREF codes, creating a collaborative library of styles. - The productized service mentioned utilizes workflow automation platforms like n8n to connect the different AI tools. These platforms can be configured to receive requests via a web form or chat, manage API calls, handle errors with fallback models, and deliver the final image, turning the multi-step process into a scalable service. - Unlike a simple filter which overlays adjustments, AI style transfer re-renders the content image using the learned stylistic elements—like brushstroke patterns, texture, and color palettes—from the reference image. - The Gemini-based workflow gives designers precise control by using a detailed text prompt to direct *how* the style reference is applied to the content image. For instance, a prompt can explicitly instruct the model to use the medium, color palette, mood, and rendering technique from the style source. - Competing tools offer different levels of control; for example, Midjourney allows users to adjust the "style weight" via a `--sw` parameter (from 0 to 1000) to fine-tune the intensity of the applied style.
