ZiT Studio - v1.0
This workflow, developed by the creator since the release of Z-Image Turbo, is designed to be versatile and not overly complex, maximizing output quality while addressing technical limitations. It includes features such as image generation, inpainting using Alibaba-PAI's ControlnetUnion-2.1, latent and tile upscaling, and the ability to use negative prompts with NAG (Noise Attention Guidance). The workflow also supports different samplers and schedulers, seed variance nodes for variety, and the use of LoRAs with ModelMerge nodes. It is organized into groups for easy toggling of image generation, inpainting, and upscaling methods. Optional features include the ZStudio LLM Prompt Enhancer and optimizations using Triton and FlashAttention.
Features & Nodes
Image generationInpaintingLatent upscaleTile upscaleUpscale with ESRGANUse of NAG for negative promptsRes4jf sampler + schedulerSeedVariance nodesModelMerge nodes for LoRAsSeparate groups for generation, inpainting, and upscalingOptional ZStudio LLM Prompt EnhancerOptional optimizations using Triton and FlashAttention