Nano AINano AI

Tongyi Lab · 2025-11-26 · 4 credits

Z-Image Turbo, Tongyi's 6B text-to-image model

Z-Image Turbo is the distilled 6B model in Tongyi-MAI's Z-Image family. The README read on 2026-11-07 says it uses 8 function evaluations, no classifier-free guidance, and Apache-2.0 weights. This form opens on Z-Image Turbo at 4 credits. It makes a PNG from a prompt. It does not edit a photo.

Best for: Low-cost rapid exploration
Try

64 free credits = 8 standard images. No card required.

Output board

Z-Image Turbo AI generated example
Z-Image Turbo AI image example

Your generated variations will appear here.

Made by
Tongyi Lab
Released
November 26, 2025
Price on Nano AI
4 credits per image
Page updated
November 7, 2026

Source: github.com

Z-Image Turbo panels from Tongyi-MAI

Cropped on 2026-11-07 from the photorealistic and bilingual showcases on the Hugging Face model card. Faces, store brands, and film-poster panels in those collages were left out. Spellings on the moss poster are theirs. These files are not new renders from this site.

Z-Image Turbo: Panel cropped from Tongyi-MAI's photorealistic showcase: a snow peak at sunrise and prayer flags, no people. Their published image, not a new Nano AI render.
Z-Image Turbo · From the official showcasePanel cropped from Tongyi-MAI's photorealistic showcase: a snow peak at sunrise and prayer flags, no people. Their published image, not a new Nano AI render.Source
Z-Image Turbo: Panel cropped from the same photorealistic showcase: fireworks over a lit arch bridge and a dark river, no people. Their published image, not a new Nano AI render.
Z-Image Turbo · From the official showcasePanel cropped from the same photorealistic showcase: fireworks over a lit arch bridge and a dark river, no people. Their published image, not a new Nano AI render.Source
Z-Image Turbo: Panel cropped from Tongyi-MAI's bilingual text showcase: a moss exhibition poster with English and Chinese titles. Small Latin names are printed as they appear, including odd spellings. Their published image, not a new Nano AI render.
Z-Image Turbo · From the official showcasePanel cropped from Tongyi-MAI's bilingual text showcase: a moss exhibition poster with English and Chinese titles. Small Latin names are printed as they appear, including odd spellings. Their published image, not a new Nano AI render.Source
Z-Image Turbo: Panel cropped from the same text showcase: a wet leaf with the Chinese line 显卡在燃烧 and the English line Glow Unobtanium X-1. Their published image, not a new Nano AI render.
Z-Image Turbo · From the official showcasePanel cropped from the same text showcase: a wet leaf with the Chinese line 显卡在燃烧 and the English line Glow Unobtanium X-1. Their published image, not a new Nano AI render.Source

What Z-Image Turbo is

Tongyi-MAI describes Z-Image as a 6B single-stream diffusion transformer family, also written 造相. The GitHub news line dated 2025-11-26 says Z-Image Turbo was released that day on Hugging Face and ModelScope. A later line, dated 2026-01-27, says the foundation checkpoint Z-Image was released then. This page is the Turbo checkpoint. The foundation model, Z-Image-Edit, and Z-Image-Omni-Base are separate rows in their model zoo.

Their model zoo, read the same day, marks Z-Image Turbo as pretrained, supervised-fine-tuned, and reinforcement-learned, with 8 steps, classifier-free guidance off, and generation only. Visual quality is marked very high and diversity low, and fine-tuning is marked not applicable. The foundation row is 50 steps with guidance on. Z-Image-Edit is marked for editing and still says to be released. Z-Image-Omni-Base is marked for generation and editing and also says to be released. This form does not load those checkpoints.

The architecture note calls the design a Scalable Single-Stream DiT, or S3-DiT: text tokens, visual semantic tokens, and image VAE tokens are one sequence. The card claims photorealistic images, English and Chinese text, and instruction following. It also claims sub-second latency on an H800 and a fit inside 16 GB of VRAM. Those are Tongyi-MAI's statements. This site has not timed a run and has not measured VRAM.

Their 2025-12-08 news line says Z-Image Turbo ranked 8th overall on the Artificial Analysis text-to-image leaderboard that day and was the top open-source model there. This page has not re-read that leaderboard on 2026-11-07, so the rank is a dated news line, not a current score. The technical report they cite is arXiv 2511.22699. The distillation notes cite Decoupled-DMD, arXiv 2511.22677, and a later note, arXiv 2511.13649.

KIE's public page for this model, read on 2026-11-07, titles it an affordable Z-Image-Turbo API by Tongyi-MAI and says the current input is text to image. It lists PNG output, prompts up to 1000 characters, and aspect ratios 1:1, 4:3, 3:4, 16:9, and 9:16. It lists a flat 0.8 KIE credits per image, about $0.004, and says new KIE users receive 80 KIE credits. Those are KIE's figures, not Nano AI credits. On this site one Z-Image Turbo image is 4 credits. Eligible new accounts receive 64 credits, which is 16 images at that price. The menu on this page opens on Z-Image Turbo. Other names in the same menu are other models this site sells, each with its own price.

Z-Image Turbo at a glance

Family
Z-Image, also 造相, from Tongyi-MAI
This checkpoint
Z-Image Turbo, distilled, generation only
Parameters
6B, from the model card
Architecture
S3-DiT, one stream for text and image tokens
Released
2025-11-26, GitHub news line
License
Apache-2.0 on the Hugging Face card
Official steps
8 NFEs; the diffusers example uses 9 steps and guidance 0
Official example size
1024 by 1024 in the diffusers snippet
Their speed claim
Sub-second on an H800; 16 GB VRAM, their words
This form
Z-Image Turbo, 4 credits, PNG, text only
Prompt length here
KIE documents a maximum of 1000 characters
Aspect ratios here
1:1, 4:3, 3:4, 16:9, 9:16
Resolution control
The documented request has no resolution field
Not this form
No edit, no LoRA loader, no ComfyUI graph
Sign-up credits
64 credits, 16 images at 4 credits

How to read a Z-Image Turbo claim

01

Turbo is the 8-step distill

The model zoo marks Z-Image Turbo at 8 steps with guidance off. The foundation Z-Image row is 50 steps with guidance on, and the news line dates that checkpoint to 2026-01-27. This form is the Turbo row.

02

Editing is a different row

Z-Image-Edit and Z-Image-Omni-Base are marked for editing and still say to be released on the zoo read on 2026-11-07. This form has no edit tab while Z-Image Turbo is selected.

03

4 credits is this site's price

KIE lists 0.8 of its own credits, about $0.004, and 80 KIE credits for a new KIE account. Nano AI charges 4 credits. 64 sign-up credits cover 16 images. Those two balances are not the same unit.

04

Speed and rank are theirs

Sub-second H800 latency, 16 GB VRAM, and the 2025-12-08 leaderboard line are Tongyi-MAI's statements. This page does not add a measured second count.

Z-Image Turbo and other image models here

Credits are Nano AI prices. The Turbo column follows the Tongyi-MAI README and KIE's page as read on 2026-11-07. The other columns are models this site already sells.

Z-Image TurboQwen Image 2.1FLUX.3 KleinNano Banana 2.1
Task on this siteText to imageText and editText and editText and edit
Credits per image4848
Images from 64 credits168168
Open weightsApache-2.0YesFLUX Kommunity LicenseNo
Edit tab on this pageHidden while Turbo is selectedAvailable on its own pageAvailable on its own pageAvailable on its own page
Vendor steps they publish8 NFEs, guidance offNot restated hereNot restated hereNot restated here

How to generate with Z-Image Turbo

  1. 1

    Write a prompt under 1000 characters

    KIE documents 1000 characters as the maximum. Describe the subject, light, and any English or Chinese words you want drawn. The four prompts below stay well under that cap.

  2. 2

    Keep Z-Image Turbo selected

    The menu opens on Z-Image Turbo at 4 credits. Text to image is the only mode for that choice. Pick 1:1, 4:3, 3:4, 16:9, or 9:16. There is no resolution dropdown because the documented request has no resolution field.

  3. 3

    Read the price, then generate

    One image is 4 credits, so 64 sign-up credits cover 16. A failed generation returns those credits under the site rules. The result is a PNG from Z-Image Turbo, not an edit of an upload.

Prompts for this Z-Image Turbo form

Written for this form. They are not the prompts inside Tongyi's showcase collage. Each one avoids people and logos.

  • Snow peak

    A snow peak at sunrise with colorful prayer flags in the foreground, cold blue shadows, no people, no readable text, no logos.

  • Night fireworks

    Fireworks over a concrete arch bridge and a dark river, city lights on the far bank, no people, no readable text, no logos.

  • Moss poster

    A flat botanical poster titled Moss Exhibition in English and 苔痕 in Chinese, ferns and a round moss drawing, cream paper, no people, no logos.

  • Wet leaf

    A single wet green leaf filling the frame, studio light, water beads on the veins, no people, no text, no logos.

FAQ

Z-Image Turbo questions

Answers use the Tongyi-MAI README, the Hugging Face card, and KIE's page as read on 2026-11-07, plus this site's credit table.

Is Z-Image Turbo the same as the foundation Z-Image?

No. The README dates Z-Image Turbo to 2025-11-26 and the foundation Z-Image checkpoint to 2026-01-27. Turbo is the 8-step distill with guidance off. The foundation row is 50 steps with guidance on. This form is Turbo.

Can this page edit an uploaded photo?

No. Z-Image Turbo is generation only. Z-Image-Edit is a separate zoo row and still said to be released on 2026-11-07. While Z-Image Turbo stays selected, the edit tab is not shown.

How many images do 64 credits buy?

Sixteen. Z-Image Turbo is 4 credits per image. Qwen Image 2.1 and Nano Banana 2.1 are 8 credits, which is 8 images from the same 64. FLUX.3 Klein is also 4 credits and can edit.

Does the form send a step count or a 1024 size?

The documented request is a prompt, an aspect ratio, and the safety check. It has no resolution field and no step field. The 1024 by 1024 size, 9 sampler steps, and guidance 0 are in Tongyi's local diffusers example, not in this form's controls.

Is the 8th-place rank still current?

The README's news line dated 2025-12-08 says that rank and a first place among open-source models. This page did not re-read the live leaderboard on 2026-11-07.

Is Nano AI affiliated with Tongyi Lab?

No. Nano AI is independent. Weights and claims above are Tongyi-MAI's. The price of 4 credits is this site's price for Z-Image Turbo.

More from Nano AI

HomeNano Banana AI image generatorGenerate and edit with every model from one prompt box.WorkbenchNano image generator and editorThe full workbench: every image model, batches and your history.PricingNano Banana price and credit plansWhat each model costs, credit packs and subscriptions.Model pageNano Banana ProGoogle's premium image model, ready to use online.

Related pages

Alibaba · LiveQwen Image 2.1Black Forest Labs · LiveFLUX.3 KleinGoogle · LiveNano Banana 2.1Meshy · GuideMeshy Image to 3D

Nano AI is an independent service and is not affiliated with Tongyi Lab or Alibaba. Family facts are from the Tongyi-MAI Z-Image README and the Hugging Face Z-Image Turbo card as read on 2026-11-07. KIE's price of 0.8 credits is quoted from kie.ai the same day and is not a Nano AI credit. This form runs Z-Image Turbo.