Z-Image Turbo - 2K Upscaler

Z-Image Turbo two-stage upscaling to 2K. Upload an image, get a sharpened, detail-recovered high-resolution output.

Most upscalers stretch pixels. This workflow adds a thinking step. Stage 1 runs RealESRGAN x4Plus, a dedicated super-resolution model that recovers fine textures and edges. Stage 2 passes the upscaled image through Z-Image Turbo's KSampler at denoise 0.33, which refines and sharpens without altering the composition. Five sampling steps with DPM++ 2M SDE. The final output is VAE-decoded and saved at approximately 2K resolution.

How does the Z-Image Turbo 2K Upscaler work?

Upload an image, optionally write a prompt to guide the refinement direction, and run. Stage 1 upscales 4x with RealESRGAN x4Plus, recovering texture and edge detail. Stage 2 runs Z-Image Turbo at denoise 0.33 for 5 steps, refining the upscaled image without changing the composition.

Input image Upload any image you want upscaled. The workflow reads the image for its resolution and content. Low-resolution photos, AI-generated images, product shots, compressed or degraded images, and game screenshots all work as input. The cleaner the source content, the sharper the final result. Heavily corrupted or blurry sources will still improve but may not reach the same quality as cleaner inputs.

Prompt (optional) The Qwen 3 4B text encoder allows prompt-guided refinement. Write a short description of what the image contains to steer the enhancement direction. This helps the diffusion pass add contextually appropriate texture: "a close-up portrait with natural skin texture," "a product shot of running shoes on a white background," "a forest landscape with detailed foliage."

Leave empty for neutral enhancement that follows the image's existing content without any directional guidance.

Stage 1: RealESRGAN x4Plus The first pass runs RealESRGAN x4Plus, which performs a clean 4x upscale and recovers fine details that compression or low resolution destroyed. This stage handles the heavy lifting: edges sharpen, micro-textures reappear, and the image reaches approximately 4x its original pixel count.

Stage 2: Z-Image Turbo refinement The 4x upscaled image feeds into Z-Image Turbo's KSampler at a denoise strength of 0.33. This low denoise value means the diffusion model refines the image rather than regenerating it. Composition, color, and structure stay locked. The 5-step DPM++ 2M SDE pass adds texture coherence and sharpness that the pure upscale step alone doesn't produce.

Denoise strength (0.33) The low denoise value is calibrated to preserve the upscaled image while adding refinement. Increasing it toward 0.5-0.6 would produce more aggressive changes to texture and detail, moving further from the source. At 0.33, the output stays true to the original while gaining the quality improvements that diffusion refinement adds.

What is the Z-Image Turbo 2K Upscaler good for?

The Z-Image Turbo 2K Upscaler is strongest for images where texture recovery and detail sharpening matter alongside resolution. Old or compressed photos, AI-generated images that need resolution bumps, product shots for print and e-commerce, and game or concept art assets all benefit from the two-stage approach over a single-model upscale.

Old and low-resolution photos. RealESRGAN recovers detail from compressed or degraded images. The diffusion refinement pass adds texture coherence that makes the result look restored rather than stretched. For family photos, scanned prints, or archival images, the two-stage approach produces results that single upscalers can't match.

AI-generated images. Take a 512px or 768px AI generation and bring it to 2K without losing the quality feel of the original. The refinement pass adds texture detail that makes the larger version look as intentional as the original rather than stretched.

Product photography for e-commerce and print. Upscale product shots to print-ready or large-format resolution while recovering fabric texture, surface detail, and edge sharpness. The prompt-guided refinement can steer the enhancement toward the specific material quality of the product.

Game screenshots and concept art. Bring in-engine screenshots or concept renders to higher resolution for portfolios, presentations, or production art. The workflow handles the kind of structured, non-photographic content that pure diffusion upscalers sometimes over-process.

Honest notes: the 0.33 denoise is calibrated not to change composition. If you need more aggressive texture generation, increase denoise carefully. Values above 0.5 start altering content rather than refining it. For heavily degraded sources, the two-stage pipeline improves quality significantly but the output quality ceiling is still limited by what the source contains.

How does the Z-Image Turbo 2K Upscaler compare to single-model upscaling?

The two-stage approach produces sharper, more texture-coherent results than RealESRGAN or a diffusion upscaler used alone. RealESRGAN recovers edges and structure. The Z-Image Turbo diffusion pass adds texture realism that interpolation-based upscalers can't generate. The result looks enhanced rather than stretched.

Single upscalers (ESRGAN variants, Lanczos interpolation) scale pixels using learned or mathematical patterns. They're fast and preserve content, but they can't add texture that wasn't there. The two-stage approach uses the diffusion model's ability to generate realistic texture at low denoise to fill in what super-resolution upscaling misses.

Compared to a diffusion-only upscaler at full denoise, the 0.33 denoise setting preserves the original image's structure and colors while still adding meaningful detail. High-denoise diffusion upscaling produces more dramatic texture but risks composition drift and hallucinated elements.

FAQ

What resolution does the Z-Image Turbo 2K Upscaler output?
Approximately 2K resolution. Stage 1 runs a 4x upscale with RealESRGAN x4Plus, then Stage 2 refines the output. The exact output resolution depends on your input: a 512px input produces approximately 2048px output after the 4x stage. Inputs closer to 512px will reach 2K; larger inputs will exceed 2K.

Does the upscaler change the composition or colors of my image?
No. The denoise strength is set to 0.33, which refines texture and sharpness without altering composition, color, or structure. The Z-Image Turbo diffusion pass adds detail to what's already there rather than regenerating the image. If you need a more aggressive transformation, increasing denoise moves it toward content change.

What does the prompt do in this workflow?
An optional prompt guides the texture refinement direction. Writing a short description of the image content helps the diffusion pass add contextually accurate texture. For a product shot, describe the product and surface. For a portrait, describe the subject and lighting. Leave it empty for neutral refinement.

What types of images work best with this workflow?
Low-resolution photos, compressed images, AI-generated images at small sizes, product shots, and game screenshots all work well. The workflow improves any input, but the quality ceiling depends on what the source contains. Heavily corrupted or motion-blurred sources will improve but produce lower final quality than cleaner inputs.

How do I run the Z-Image Turbo 2K Upscaler online?
You can run this workflow online through Floyo. No installation, no setup. Open the workflow in your browser, upload your image, and hit run. Free to try.