模型介紹為中英雙語，中文使用者可以關閉翻譯。
Trubo:
Pornmaster Z Image Turbo Workflow
Pornmaster Z Image Turbo Controlnet Workflow
PornMaster Z-image-turbo Controlnet: Cosplay - Anime to Realistic Style
PornMaster Z-image Increase realism & detail & sharpness
Pornmaster Z-Image Turbo i2i Workflow
PornMaster Simple 4X Magnification Workflow
Pornmaster Z-Image Turbo T2I Workflow-Double checkpoints & realism enhancer
--
Base:
Pornmaster Z Image Base t2i Workflow
--
Stable Diffusion WebUI Forge - Neo
--
如果使用nsfw的lora時，lora權重從0.1開始，然後逐漸增加調整權重。
(可以使用"Z-Image base"的lora，但是需要較高的權重。)
提高照片的解析度可以增加生成良好陰莖和陰部的成功率。
注意:有些人訓練的角色lora可以使用，有些人訓練的則無法使用，我並不知道原因。
如果對生成的性器官不滿意請使用PornMaster SDXL 或PornMaster flux2 klein 9b base 的Inpainting 檢查站來修復性器官。
--
When using nsfw's LoRa, start the LoRa weights from 0.1 and gradually increase them.
(You can use LoRa with a "Z-Image base", but it requires higher weights.)
Increasing the image resolution increases the success rate of generating good-looking penises and genitals.
Note: Some people's character lora training works, while others' does not; I don't know why.
If you are not satisfied with the generated sex organs, please use the Inpainting checkpoint of PornMaster SDXL or PornMaster flux2 klein 9b base to repair the sex organs.
--
Clip: qwen_3_4b.safetensors
Vae: ae.safetensors
--
我已將模型上傳至runninghub。
I have uploaded the model to runninghub.
PornMaster z-image Turbo V0.1_fp8:
https://www.runninghub.ai/model/public/2012876652145020930/?inviteCode=rh-v1304
PornMaster z-image Turbo V0.1_bf16:
https://www.runninghub.ai/model/public/2012891520348725249/?inviteCode=rh-v1304
PornMaster z-image Turbo V0.2_fp8:
https://www.runninghub.ai/model/public/2012870272323424257/?inviteCode=rh-v1304
PornMaster z-image Turbo V0.2_bf16:
https://www.runninghub.ai/model/public/2012858316384374786/?inviteCode=rh-v1304
PornMaster z-image Turbo V1_fp8:
https://www.runninghub.ai/model/public/2015385918403780610/?inviteCode=rh-v1304
PornMaster z-image Turbo V1_bf16:
https://www.runninghub.ai/model/public/2015380234719207425/?inviteCode=rh-v1304
PornMaster z-image Turbo V2_fp8:
https://www.runninghub.ai/model/public/2023636060722434049/?inviteCode=rh-v1304
PornMaster z-image Turbo V2_bf16
https://www.runninghub.ai/model/public/2023630811089604610/?inviteCode=rh-v1304
PornMaster_z-image_base_V0.1_fp8:
https://www.runninghub.ai/model/public/2024107278651105282/?inviteCode=rh-v1304
PornMaster_z-image_base_V0.1_bf16:
https://www.runninghub.ai/model/public/2024107273051709442/?inviteCode=rh-v1304
PornMaster_z-image_base_V0.1_distilled_fp8
https://www.runninghub.ai/model/public/2025833520878325762/?inviteCode=rh-v1304
PornMaster_z-image_base_V0.1_distilled_bf16
https://www.runninghub.ai/model/public/2025836003751108609/?inviteCode=rh-v1304
PornMaster z-image Turbo_5steps_V1_fp8
https://www.runninghub.ai/model/public/2026332244041801729/?inviteCode=rh-v1304
PornMaster z-image Turbo_5steps_V1_bf16
https://www.runninghub.ai/model/public/2026340550038786050/?inviteCode=rh-v1304
PornMaster z-image Turbo V3_fp8
https://www.runninghub.ai/model/public/2040397408881876993/?inviteCode=rh-v1304
PornMaster z-image Turbo V3.5_fp8
https://www.runninghub.ai/model/public/2041355991928807426/?inviteCode=rh-v1304
PornMaster z-image Turbo V3.5_fp16
https://www.runninghub.ai/model/public/2049373402028318721/?inviteCode=rh-v1304
--
不要問我如何安裝的問題與comfyui如何使用。
客製化lora:我的Telegram ID:@iamddtla0620
我的微信ID:iamddtla
我的line ID:iamddtla
時間寶貴，請直接說明來意，例如:訂製lora.
Do not ask me questions about how to install it or how to use ComfyUI.
Customized lora: My Telegram ID: @iamddtla0620
My WeChat ID: iamddtla
My Line ID: iamddtla
Time is precious, please state your intention directly, for example: customized LORA.
Description
與v0.1版本相比，能夠相容更多的lora，並且能夠好的還原檢查站原本內建的人物。
如果使用nsfw的lora時，lora權重從0.1開始，然後逐漸增加調整權重。
這是一個測試版本，不要對它有太高的期待，nsfw的姿勢和性器官有些微改進，如果性器官要像SDXL版本那樣良好，還有一段很長的路要走。
提高照片的解析度可以增加生成良好陰莖和陰部的成功率。
如果對生成的性器官不滿意請使用PornMaster SDXL系列的Inpainting 檢查站來修復性器官。
Compared to version 0.1, it is compatible with more LoRa characters and better restores the original built-in characters at checkpoints.
When using the NSFW LoRa, the LoRa weight starts at 0.1 and is gradually increased to adjust the weight.
This is a test version, so don't have too high expectations. The positions and genitals in nsfw have been slightly improved, but there is still a long way to go before the genitals are as good as the SDXL version.
Increasing the resolution of the photo can increase the success rate of generating good-looking penises and genitals.
If you are not satisfied with the generated sex organs, please use the Inpainting checkpoint of the PornMaster SDXL series to repair them.
FAQ
Comments (52)
V2F8版本生出图片是一堆马赛克，后台提示unet missing: ['norm_final.weight']和关于noise和layers的unet unexpected提示。请问是需要升级其他辅助软件的版本吗？
comfyui嗎?
@iamddtla 是的，v0.3.77
@Leekamlong 那你應該先測試官方原版的z-image，看能不能用，如果能用的話，我的版本應該也沒問題.
The FP8 version is the first z-image model that I haven't been able to get anything other than noise (no errors) with in Forge Neo. I tried multiple AEs and text encoders, including the ones mentioned in the description. I also tried multiple sampler/scheduler combinations (e.g., Euler / Simple), to no avail. If anyone has working setting for Forge Neo, please share them.
I tested v0.2_fp8, and it doesn't work on "Stable Diffusion WebUI Forge - Neo," for reasons I don't know, but it works fine on comfyui.
However, v0.2_bf16 works on "Stable Diffusion WebUI Forge - Neo."
Thanks for the quick response! BF16 variant doesn't play with my VRAM. So, I'm hoping that future FP8 versions don't have this issue. Your SDXL model is one of my favorites. Keep up the good work!
@kepem If you use v0.2_fp8 on comfyui, the error will not be displayed.
@iamddtla Sure, but I want to use Forge Neo rather than ComfyUI. I'm a perfectionist and prefer the simplicity of Forge Neo. I understand that ComfyUI is more flexible and powerful overall, but it's also a rabbit hole. My brain is not wired to just use a workflow that someone else set up and not tinker with it for many, many hours. ;)
超級感謝！這是我最愛的Z-Image model！這model用起來速度比原生的更快，且不會太改變太多原Z-Image裡本來訓練的角色。請問有什麼祕訣做到這樣的嗎？
我只是簡單的合併模型而已.
@iamddtla 謝謝回覆 :) 不過速度變快這部份是蠻令我感到驚訝開心的 (20 sec. -> 11 sec., 4060 16GB, 9 steps) ，所以才那麼好奇如何達成。
@LittleTurty 你使用FP8版本嗎?
@iamddtla 我是使用fp8版本，但我也是跟原生轉fp8版的比，你的模式速度就快了一倍 :)
@LittleTurty 我也不知道它生成圖片的速度為何會變快.
@iamddtla 謝謝回覆。我等能下載bf16時再試試。看看是您bf16->fp8時造成的影響(變快)、或者是您在合併模型時就變快的，很好奇 :) 再次感謝
我試了V0.1 bf16, v0.1 fp8, v0.2 fp8，在我的電腦上速度約是 14.5: 13.5: 9；v0.1已比原生的Z-image快、v0.2比v0.1又更快，而fp8比bf16快，供參考，謝謝。
(v0.2) 試了bf16，大概都比fp8時間多了1.5倍，不過還是比原生的快 :) 期待可以下載v1.0試試的那天 :)
今天試了v1，不過已無v0.2 fp8神奇的快速生成的特性，可能當時v0.2合併有些不預期的狀況造成現象沒有再重現。因電腦硬體效果不高，我之後我可能會先用v0.2，有喜歡的圖再用v1 :)
@LittleTurty v0.2 fp8版本有經過gpt提供的量化方式，但會導致無法再forge上使用，所以v1我就沒再使用相同的方式，v0.2 fp8雖然速度較快，但也算是某種"故障"，如果不使用forge倒是完全沒關係。 順便一提我的顯示卡記憶體只有12GB.
@iamddtla 感謝詳細回覆。原來是gpt提供的量化方式造成的。請問有沒有可能指導(付費)如何進行這種量化 (4060ti 16gb可以嗎)，想說可以幫自己省下不少出圖時間 :)
@LittleTurty 直接問GPT就好了吧? 我要製作z image檢查站的fp8版本，參考下面兩個網址告訴我該如何製作，https://github.com/silveroxides/convert_to_quant , https://files.catbox.moe/nzgm9s.txt
@iamddtla 謝謝回覆，我剛是直接問了gemini。現在有了你這兩個網址，應該更沒問題，我等會再問問chatgpt，從頭學起做起，再次感謝。(萬事起頭難，我是python小白 :) )
The best NSFW model for now, thanks for your great work!
v0.2FP8生产了一堆马赛克 用的是你的工作流，官方版本我是可以正常生图的，comfyui
蠻奇怪的，v0.2_fp8我在comfyui能正常使用
尝试下更新comfyui，cuda toolkit和pytorch的版本，我之前也是马赛克，更新完就没问题了
fp4 for blackwell please.
I obtained excellent results with this fp8 model using: 12 steps, res_multistep, and sgm_uniform. In ComfyUI.
Could you post example images?
I got only messy noise
Stable Diffusion WebUI Forge - Neo?
@iamddtla No, latest ComfyUI with workflow from sampled images
@210881175 Pornmaster Z Image Turbo Workflow
@iamddtla same result, see picture
@210881175 Please download a checkpoint for Z-Image Turbo created by someone else and test it to see if the problem is with your computer system.
@iamddtla There are no such problems with other checkpoints.
@210881175 same here but I do have forge neo, another fp8 works well
I have comfy, but it's impossible to load this checkpoint. In UNET? No. With classic CKPTs? No. ???
@iamddtla Of course I used this workflow, it doesn't work
Bro has "Om" profile and using Pornmaster XD
@6vidit9 which folder does one have to place this model at ? lora folder? checkpoints? Diffusion models?
Love your SDXL models but this seems hmm, maybe because its early?
Big thing I don't seem to like is this model seems to have lost the 'realism' and grain that the original model has, this looks a lot more like past base models in terms of artificialness or am i missing something? Switching between this and the base model with extra lora's, the latter for me gives better results but I'm sure I must be missing something, is there a list of terms this has been trained on or anything?
Thanks
have anyone figure a work around with forge neo?
I think I made a mistake when quantizing to fp8, which is why it can't be used on "forge neo". I will try to fix this problem when I make the next new version.
@iamddtla oh that's very kind, no hurry
@iamddtla That means the next version is work in progress good to hear.
@renderguy Version V1 has fixed the issue that it could no longer be used with Forge.
@iamddtla Awesome!! Thanks! Oh...
Speaking of Forge Neo, i just started using it today. And I dont see the option to select Z image there. Just SDXL,FLUX,WAN etc etc. Do I need to update it or just select SDXL?
@KaiserWilhelm qwen.