Universal ComfyUI Krea 2 Studio
The workflow’s defining feature is its coordinated three-stage generation technique, designed to build high-resolution photographs with rich detail and natural texture directly within one generation chain.
Establish the image. The first sampler creates composition, lighting and the main subject at a modest resolution.
Develop the detail. Model-assisted upscaling increases the working resolution. A second sampling pass develops detail at that larger size, followed by RTX size adjustment.
Refine and finish. A third sampling pass refines the image before the final 2× VAE decode delivers the enlarged output.
Detail develops through repeated diffusion sampling as resolution increases. Step 1 always runs; Steps 2 and 3 can be enabled independently. Step 3 automatically adjusts its sampling parameters according to whether Step 2 ran.
The interface is designed to explain the process. Clear functional groups, readable node names, logically ordered sockets and labelled connections make each operation easy to locate. Routing follows the flow of work, with deliberate spacing and minimal crossings. Spaghetti solutions are rigorously avoided. Subgraphs organize complexity while keeping the main controls accessible.
Manual prompting, local LM Studio editing, Florence image description, LoRA selection, optional eye refinement and image comparison support the central generation process.
The accompanying dependency note covers installation and model files. Models are not bundled; RTX processing requires an NVIDIA RTX GPU.
Licence: Copyright © 2026 BlandiZ. Original contributions are free for personal and commercial use under MIT; the full licence is included in the workflow JSON. Third-party components retain their own terms. Supplied as is. Responsibility for inputs, outputs and their use rests with the operator; misuse is not endorsed. Liability exclusions apply only where permitted by law.