7 ms·
Z-Image: Powerful and highly efficient image generation model with 6B parameters
- Copenjin 11mo agoVery good, not always perfect with text or with following exactly the prompt, but 6B so... impressive.
- accrual 11mo agoI have had good textual results with the Turbo version so far. Sometimes it drops a letter in the output, but most of the time it adheres well to both the text requested and the style. I tried this prompt on my username: "A painted UFO abducts the graffiti text "Accrual" painted on the side of a rusty bridge." Results: https://imgur.com/a/z-image-test-hL1ACLd https://imgur.com/a/z-image-test-hL1ACLd
- pawelduda 11mo agoDid anyone test it on 5090? I saw some 30xx reports and it seemed very fast
- Wowfunhappy 11mo agoEven on my 4080 it's extremely fast, it takes ~15 seconds per image.
- accrual 11mo agoDid you use PyTorch Native or Diffusers Inference? I couldn't get the former working yet so I used Diffusers, but it's terribly slow on my 4080 (4 min/image). Trying again with PyTorch now, seems like Diffusers is expected to be slow.
- Wowfunhappy 11mo agoUh, not sure? I downloaded the portable build of ComfyUI and ran the CUDA-specific batch file it comes with. (I'm not used to using Windows and I don't know how to do anything complicated on that OS. Unfortunately, the computer with the big GPU also runs Windows.)
- accrual 11mo agoHaha, I know how it goes. Thanks, I'll give that a try! Update: works great and much faster via ComfyUI + the provided workflow file.
- egeres 11mo agoIncredibly fast, on my 5090 with CUDA 13 (& the latest diffusers, xformers, transformers, etc...), 9 samplig steps and the "Tongyi-MAI/Z-Image-Turbo" model I get: - 1.5s to generate an image at 512x512 - 3.5s to generate an image at 1024x1024 - 26.s to generate an image at 2048x2048 It uses almost all the 32Gb Gb of VRAM and GPU usage. I'm using the script from the HF post: https://huggingface.co/Tongyi-MAI/Z-Image-Turbo https://huggingface.co/Tongyi-MAI/Z-Image-Turbo
- SV_BubbleTime 11mo agoWeird, even at 2048 I don’t think it should be using all your 32GB VRAM.
- egeres 11mo agoIt stays around 26Gb at 512x512. I still haven't profiled the execution or looked much into the details of the architecture but I would assume it trades off memory for speed by creating caches for each inference step
- SV_BubbleTime 11mo agoIDK. Seems odd. It’s an 11GB model, I don’t know what it could caching in ram.
- danielbln 11mo agoWe've come a long way with these image models, and the things you can do with paltry 6B are super impressive. The community has adopted this model wholesale, and left Flux(2) by the way side. It helps that Z-Image isn't censored, whereas BFL (makers of Flux 2) dedicated like a fith of their press release talking about how "safe" (read: censored and lobotomized) their model is.
- rfoo 11mo agoBut this is a CCP model, would it refuse to generate Xi?
- vunderba 11mo agoYou tell me. https://imgur.com/a/7FR3uT1 https://imgur.com/a/7FR3uT1
- CamperBob2 11mo agoIt will generate anything. Xi/Pooh porn, Taylor Swift getting squashed by a tank at Tiananmen Square, whatever, no censorship at all. With simplistic prompts, you quickly conclude that the small model size is the only limitation. Once you realize how good it is with detailed prompts, though, you find that you can get a lot more diversity out of it than you initially thought you could. Absolute game-changer of a model IMO. It is competitive with Nano Banana Pro in some respects, and that's saying something.
- cubefox 11mo agoI could imagine the Chinese government is not terribly interested in enforcing its censorship laws when this would conflict with boosting Chinese AI. Overregulation can be a significant inhibitor to innovation and competitiveness, as we often see in Europe.
- CamperBob2 11mo agoI'm sure they're also aware that few of their own citizens are in a position to run the model themselves, and that it's easy enough to use the system prompt to censor hosted copies for domestic consumption. Censoring open-source models really doesn't make a lot of sense for China. Which could also be why local Deepseek instances are relatively easy to jailbreak.
- xnx 11mo agoZ-Image seems to be the first successor to Stable Diffusion 1.5 that delivers better quality, capability, and extensibility across the board in an open model that can feasibly run locally. Excitement is high and an ecosystem is forming fast.
- SV_BubbleTime 11mo agoDid you forget about SDXL? Clearly you have, but while on the topic, it is amazing to me that only came out 2.5 years ago.
- razster 11mo agoI’ve paired my Z-Image Turbo with SeedVR2 upscale, running on a RTX3060 12gb, 32gb sysMEM, generates in 40sec. I’m holding out for Z-Image Edit that is a larger model, once that is out… going to be interesting. Oh and to train your own ZIT LoRA, takes 5hrs for 3000 steps. So fast.
- dragonwriter 11mo agoZ-Image Base and Z-Image Edit have been announced as being the same size (or, at least, the whole set has been announced as being in the 6B size class) as Turbo, but slower (50 steps with CFG, apparently, from the announced 100 NFEs compared to Turbo's 9 NFEs, where turbo doesn't, in the use they reference, use CFG.)
- vunderba 11mo agoI've done some preliminary testing with Z-Image Turbo in the past week. Thoughts - It's fast (~3 seconds on my RTX 4090) - Surprisingly capable of maintaining image integrity even at high resolutions (1536x1024, sometimes 2048x2048) - The adherence is impressive for a 6B parameter model Some tests (2 / 4 passed): https://imgpb.com/exMoQ https://imgpb.com/exMoQ Personally I find it works better as a refiner model downstream of Qwen-Image 20b which has significantly better prompt understanding but has an unnatural "smoothness" to its generated images.
- echelon 11mo agoSo does this finally replace SDXL? Is Flux 1/2/Kontext left in the dust by the Z Image and Qwen combo?
- tripplyons 11mo agoSDXL has been outclassed for a while, especially since Flux came out.
- aeon_ai 11mo agoSubjective. Most in creative industries regularly still use SDXL. Once Z-image base comes out and some real tuning can be done, I think it has a chance of replacing it for the function SDXL has
- Scrapemist 11mo agoSource?
- echelon 11mo agoMost of the people I know doing local AI prefer SDXL to Flux. Lots of people are still using SDXL, even today. Flux has largely been met with a collective yawn. The only thing Flux had going for it was photorealism and prompt adherence. But the skin and jaws of the humans it generated looked weird, it was difficult to fine tune, and the licensing was weird. Furthermore, Flux never had good aesthetics. It always felt plain. Nobody doing anime or cartoons used Flux. SDXL continues to shine here. People doing photoreal kept using Midjourney.
- zkmon 11mo agoJust want to learn - who actually needs or buys up generated images?
- nine_k 11mo agoSome ideas for your consideration: - Illustrating blog posts, articles, etc. - A creativity tool for kids (and adults; consider memes). - Generating ads. (Consider artisan production and specialized venues.) - Generating assets for games and similar, such as backdrops and textures. Like any tool, it takes certain skill to use, and the ability to understand the results.
- zkmon 11mo agoExcept for gaming, that doesn't sound like a huge market worthy of pouring millions into training these high-quality models. And there is a lot of competition too. I suspect there are some other deep-pocketed customers for these images. Probably animations? movies? TV ads?
- pixl97 11mo agoPropaganda?
- nine_k 11mo agoI'd say that picture ad market alone would suffice. OTOH these are open-weight models released to the public. We don't get to use more advanced models for free; the free models are likely a byproduct of producing more advanced models anyway. These models can be the freemium tier, or gateway drugs, or a way of torpedoing the competition, if you don't want to believe in the goodwill of their producers.
- Zopieux 11mo ago>A creativity tool for kids (and adults; consider memes). Fixed that for you: (and adults; consider porn). I don't think you realize the extent of the “underground” nsfw genai community, which has to rely on open-weight models since API models all have prude filters.
- nine_k 11mo agoIt's amazing how much knowledge about the world fits into 16 GiB of the distilled model.
- echelon 11mo agoThis is early days, too. We're probably going to get better at this across more domains. Local AI will eventually be booming. It'll be more configurable, adaptable, hackable. "Free". And private. Crude APIs can only get you so far. I'm in favor of intelligent models like Nano Banana over ComfyUI messes (the future is the model, not the node graph). I still think we need the ability to inject control layers and have full access to the model, because we lose too much utility by not having it. I think we'll eventually get Nano Banana Pro smarts slimmed down and running on a local machine.
- bobsmooth 11mo ago>Local AI will eventually be booming. With how expensive RAM currently is, I doubt it.
- api 11mo agoI’m old enough to remember many memory price spikes.
- SV_BubbleTime 11mo agoI remember saving up for my first 128MB stick and the next week it was like triple in price.
- lomase 11mo agoDo you also remember when eveybody was waiting for cryto to cool off to buy a GPU?
- echelon 11mo agoIt's temporary. Sam Altman booked all the supply for a year. Give it time to unwind.
- muglug 11mo agoThe [demo PDF](https://github.com/Tongyi-MAI/Z-Image/blob/main/assets/Z-Image-Gallery.pdf https://github.com/Tongyi-MAI/Z-Image/blob/main/assets/Z-Ima...) has ~50 photos of attractive young women sitting/standing alone, and exactly two photos featuring young attractive men on their own. It's incredibly clear who the devs assume the target market is.
- abbycurtis33 11mo agoThey're correct. This tech, like much before it, is being driven by the base desires of extremely smart young men.
- IncreasePosts 11mo ago[flagged]
- pixl97 11mo agoI mean spending all that time on dates, and wives, and kids gives you much less time to build AI models. The people with the time and desire to do something are the ones most likely to do it, this is no brilliant observation.
- IncreasePosts 11mo agoYou could say that about any field, and yet we don't see the same behavior in most other fields Spending all your time on dates and wives and kids means you're not spending all your time building houses.
- pixl97 11mo agoI mean things that take hard physical labor are typically self limiting... I do nerdy computer things and I actually build things too, for example I busted up the limestone in my backyard in put in a patio and raised garden. Working 16 hours a day coding/or otherwise computering isn't that hard even if your brain is melted at the end of the day. 8 - 10 of physically hard labor and your body starts taking damage if you keep it up too long. And really building houses is a terrible example! In the US we've been chronically behind on building millions of units of houses. People complain the processes are terribly slow and there is tons of downtime. So yea, I don't think your analogy works at all.
- khimaros 11mo agoi have been testing this on my Framework Desktop. ComfyUI generally causes an amdgpu kernel fault after about 40 steps (across multiple prompts), so i spent a few hours building a workaround here https://github.com/comfyanonymous/ComfyUI/pull/11143 https://github.com/comfyanonymous/ComfyUI/pull/11143 overall it's fun and impressive. decent results using LoRA. you can achieve good looking results with as few as 8 inference steps, which takes 15-20 seconds on a Strix Halo. i also created a llama.cpp inherence custom node for prompt enhancement which has been helping with overall output quality.
- xfalcox 11mo agoWe have vLLM for running text LLMs in production. What is the equivalent for this model?
- mh- 11mo agoI would say there's isn't an equivalent. Some people will probably tell you ComfyUI - you can expose workflows via API endpoints and parameterize them. This is how e.g. Krita AI Diffusion uses a ComfyUI backend. For various reasons, I doubt there are any large scale SaaS-style providers operating this in production today.
- salty_frog 11mo agoI'm intrigued by the various reasons why you think there are not any large scale SAAS operating this in production?
- threeebo 11mo agoi dont believe there is a viable use case for large scale AI-generated images as there is for text... except for porn, but many orgs with SAAS capabilities wouldn't touch that
- mh- 11mo agoI was referring to attributes of the ComfyUI software/project, not the idea of serving an image generation API to be clear. There are several of those providers.
- idontwantthis 11mo agoDoes it run on apple silicon?
- iamflimflam1 11mo agoIt's working for me - it does max out my 64GB though.
- sheepscreek 11mo agoWow. I always forget how unlike autoregressive models, diffusion models are heavier on resources (for the same number of parameters).
- sheepscreek 11mo agoApparently - https://github.com/ivanfioravanti/z-image-mps https://github.com/ivanfioravanti/z-image-mps Supports MPS (Metal Performance Shaders). Using something that skips Python entirely along with a mlx or gguf converted model file (if one exists) will likely be even faster.
- opensandwich 11mo ago(Not tested) though apparently it already exists: https://github.com/leejet/stable-diffusion.cpp/wiki/How-to-Use-Z%E2%80%90Image-on-a-GPU-with-Only-4GB-VRAM https://github.com/leejet/stable-diffusion.cpp/wiki/How-to-U...
- BoredPositron 11mo agoI wish they would have used the WAN vae.
- thih9 11mo agoAs an AI outsider with a recent 24GB macbook, can I follow the quick start[1] steps from the repo and expect decent results? How much time would it take to generate a single medium quality image? [1]: https://github.com/Tongyi-MAI/Z-Image?tab=readme-ov-file#-quick-start https://github.com/Tongyi-MAI/Z-Image?tab=readme-ov-file#-qu...
- altmanaltman 11mo agoIf you don't know anything about AI in terms of how these models are run, comfyui's macos version is probably the easiset to use. There is already a Z-Image workflow that you can get and comfyui will get all the models you need and get it work together. Can expect decent speed
- thih9 11mo agoI'm fine with the quick start steps and I prefer CLI to GUI anyway. But if I try it and find it too complex, I now know what to try instead - thanks. I'm still curious whether this would run on a MacBook and how long would it take to generate an image. What machine are you using?
- egeozcan 11mo agoHave a 48GB M4 Pro and every inference step takes like 10 seconds on a 1024x1024 image. so six steps and you need a minute. Not terrible, not great.
- aleyan 11mo agoI have a 24GB M5 macbook pro. In ComfyUI using default z-image workflow, generating a single image just took me 399 seconds, during which the computer froze and my airpods lost audio. On replicate.com a single image takes 1.5s at a price of 1000 images per $1. Would be interesting to see how quick it is on ComfyUI Cloud. Overall, running generative models locally on Macs seems very poor time investment.
- airtonix 11mo ago[dead]
- cubefox 11mo agoI'm particularly impressed by the fact that they seem to aim for photorealism rather than the semi-realistic AI-look that is common in many text-to-image models.
- CamperBob2 11mo agoExactly, and at the same time, if you want an affected style, all you have to do is ask for it.
- bilsbie 11mo agoWhat kind of rig is required to run this?
- CamperBob2 11mo agoThe simple Python example program runs great on almost any GPU with 8 GB or more memory. Takes about 1.5 seconds per iteration on a 4090. The bang:buck ratio of Z-Image Turbo is just bonkers.
- b0ner_t0ner 11mo agoCPU can be used: https://github.com/rupeshs/fastsdcpu/pull/346 https://github.com/rupeshs/fastsdcpu/pull/346
- reactordev 11mo agoMy issue with this model is it keeps producing Chinese people and Chinese text. I have to very specifically go out of my way to say what kind of race they are. If I say “A man”, it’s fine. A black man, no problem. It’s when I add context and instructions is just seems to want to go with some Chinese man. Which is fine, but I would like to see more variety of people it’s trained on to create more diverse images. For non-people it’s amazingly good.
- orbital-decay 11mo agoAll modern models have their default looks. Meaningful variety of outputs for the same inputs in finetuned models is still an open technical problem. It's not impossible, but not solved either.
- SV_BubbleTime 11mo agoI’m not sure how this is anything but a plus. It means it respects nationality choices and if you don’t mention it that is your bad prompting and not a failure to not have the default nationality you would prefer.
- ForOldHack 11mo agoIt would be more useful to have some standards on what one could expect in terms of hardware requirements and expected performance.
- thot_experiment 11mo agoI've messed with this a bit and the distill is incredibly overbaked. Curious to see the capabilities of the full model but I suspect even the base model is quite collapsed.
- gatane 11mo agoDude, please give money to artists instead of using genAI
- phantomathkg 11mo agoUnfortunately, another China censored model. Simply ask it to generate "Tank Man" or "Lady Liberty Hong Kong" and the model return a blackboard with text saying "Maybe Not Safe".
- user34283 11mo agoThis is an issue with your provider. You need to download the model. It generates an image of a tank and the statue of liberty for those prompts.
- icyfox 11mo agoWe talked about this model in some depth on the last Pretrained episode: https://youtu.be/5weFerGhO84?si=Eh_92_9PPKyiTU_h&t=1743 https://youtu.be/5weFerGhO84?si=Eh_92_9PPKyiTU_h&t=1743 Some interesting takeaways imo: - Uses existing model backbones for text encoding & semantic tokens (why reinvent the wheel if you don't need to?) - Trains on a whole lot of synthetic captions of different lengths, ostensibly generated using some existing vision LLM - Solid text generation support is facilitated by training on all OCR'd text from the ground truth image. This seems to match how Nano Banana Pro got so good as well; I've seen its thinking tokens sketch out exactly what text to say in the image before it renders.
- GuestFAUniverse 11mo agoAll the examples I tried were garbage. Looked decent -- no horrors -- but didn't do the job. Anything with "most cultures" were manga-influenced comic strips with kanji. Useless.
- GaggiX 11mo ago>manga-influenced comic strips with kanji. Useless. Are you sure it was Japanese? Because the model is Chinese so it's likely to output Chinese (it happened in my testing).
- GuestFAUniverse 11mo agoHonestly I don't know if it was (Simplified) Chinese, or Japanese Kanji (so, symbols derived from Chinese). And it isn't even relevant. "most cultures" cannot read anything of it. So what's the nitpicking about?
- GaggiX 11mo ago>So what's the nitpicking about? Idk I just thought it was funny to read the ignorant comment that called the Chinese model useless because it rendered Chinese text and calling it Japanese. The model is trained to render English or Chinese text.
- Tepix 11mo agoI‘m wondering: Is it faster or slower when spread across two GPUs (RTX3090)?
- ArcaneMoose 11mo agoThis model is awesome. I am building an infinite CYOA game and this was a drop-in replacement for my scene image generation. Faster, cheaper, and higher quality than what I was using before!
- baby6343 11mo ago[dead]