Mastodawn

Der Elektroheiland

Prompt: fraudulent useless bloated overengineered kludged together dysfunctional artificial intelligent machine posing as the saviour of humanity and herald of a bright science fiction future while almost falling apart, #surrealism #dada #elektrodada fantastic dreamlike magic realism Magritte Dali Yerka Pranckevicius

#aiart #degenerativeart #zimage #zimageturbo

PsychoticSheep Feb 8

These images were created with Z-Image-Turbo (FP8) in ComfyUI

Z-Image differs from SDXL primarily in its architecture and text processing. While SDXL uses two CLIP text encoders, Z-Image works with Qwen as a text encoder, which means text and image information are processed directly in a single transformer stream. This makes Z-Image particularly strong at prompt understanding, even for longer or multilingual texts.

The exciting part is that Z-Image can generate images extremely quickly, often in just a few steps, and delivers consistent, clean results. In contrast, SDXL focuses more on maximum detail, complex scenes, and flexible control. Z-Image demonstrates how efficient architectures and specialized text encoders can change AI image generation.

#ZImageTurbo #AIArt #AIGenerated #QwenTextEncoder #DigitalArt #AIArtCommunity #CreativeAI #TextToImage #FastAI #ArtGeneration #MachineLearning #AIDesign #AIExperiment #LocalAi

Dr. Fortyseven 🥃 █▓▒░Jan 20

#aiart #zimageturbo #whendarknessmeetsthelight

Dr. Fortyseven 🥃 █▓▒░Jan 15

One of these photos is AI generated using the Z-Image Turbo model (running on local hardware), and the other is a real photo taken in Minnesota.

I've placed the answer in the alt text.

If you'd asked me which one was the generated image, I'd have probably chosen the real one.

Maybe, MAYBE with meticulous, slow examination I could have figured it out. Or at least attributed a chance percentage of it being generated.

But man... we've really fucked ourselves, I think. This is just the state of it all TODAY. And that's just here at home on consumer hardware. And this isn't even the best, but where the hell else is there to go in terms of image fidelity for uses like this?

And it comes at the absolute worst time in US history, when video and photograph evidence is the most crucial...

#genai #zimageTurbo #ai

sayzard Jan 15

Janek Mann (@janekm)

작성자는 논문/발표자의 벤치마크 결과를 이해하지 못하겠다며, 벤치마킹을 심하게 잘못했거나 공개된 모델에 버그가 있을 수 있다고 지적합니다. 또한 Z-Image-Turbo(두번째 이미지)가 본인 환경에서 더 나은 텍스트 출력을 보여준다고 언급해 출시된 모델의 성능/재현성 문제를 제기하고 있습니다.

https://x.com/janekm/status/2011415256506179735

#imagemodels #benchmarking #zimageturbo #modelbug

Janek Mann (@janekm) on X

@cedric_chee Yeah, I also don't understand their benchmark results... Either they messed up bad in their benchmarking or the released model has a bug. Even Z-Image-Turbo (2nd image) is giving me better text output 😅

X (formerly Twitter)