Z-Image Turbo: enhanced text encoder + Flow-DPO + Ultra Real, tested at a fixed seed
Three additions to Z-Image Turbo, tested properly: same prompt, same seed, one variable at a time. The results are unusually clear-cut for this genre. The fine-tuned text encoder makes objects that were missing appear. Flow-DPO improves colour and light. Ultra Real, at full weight, degrades the image — and the author says so instead of hiding it. Evidence note: both comparison graphs ship as JSON…
Three enhancements were tested for Z-Image Turbo, each tested one variable at a time. The fine-tuned text encoder, built on the Qwen3 architecture, aims to improve visual narrative rather than simple keyword expansion. Flow-DPO improves colour and lighting, while Ultra Real, at full strength, degrades the image. Both comparison graphs are provided as JSON files.
The 9-step CFG-1 sampler, model and encoder filenames, LoRA names, and weights used in each variant are listed. The article states that enhanced text encoder can make missing objects appear, Flow-DPO improves colour and light, and Ultra Real degrades the image. The author recommends using the enhanced encoder and Flow-DPO together, with Ultra Real at a reduced weight.
Inside ControlNet segmented rendering, the enhanced encoder breaks pose control, but this can be fixed by using the high-noise segment to lock the shape and the low-noise segment to add detail. Both graphs are based on a single prompt and seed, and the sample size is small.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
