zimage far better than this glm-image
prompt:
A cinematic studio portrait photo of a smiling young emma watson, half-body shot, wearing no [outfit], holding a tiny 3D cartoon mini character version of themselves between fingers (pinching the mini figure gently). The mini character is Pixar/Disney-style 3D, big expressive eyes, cute proportions, wearing matching outfit colors, happy pose with raised hands. The real person is ultra-realistic photography, natural skin texture, detailed hair, soft smile. Warm golden background with strong bokeh lights, dreamy glow, soft studio lighting, shallow depth of field, premium commercial look, 85mm lens, f/1.8, high resolution, sharp focus on face and mini character, clean composition, realistic shadows and reflections.
glm-image result:
You want to share a comparison, fine, but at least adopt a scientific approach, be impartial in your title.
You are on the official model repo, not on Reddit here. Have some respect for the people who have spent their time making it available for you, for free.
Posting a crude title and simple prompt with images is not the way to go about starting a discussion on here. It is better to be more constructive and in depth with your findings while being respectful to hard working developers/engineers. Common decency please.
AFAIK, auto-regressive image generation is still a somewhat experimental thing, and they did not release this model and claim it is the best open source image model or something.
Reduce your porn viewing, you stupid teenager. lol
Get a refund then 😮💨
prompt:
A cinematic studio portrait photo of a smiling young emma watson, half-body shot, wearing no [outfit], holding a tiny 3D cartoon mini character version of themselves between fingers (pinching the mini figure gently). The mini character is Pixar/Disney-style 3D, big expressive eyes, cute proportions, wearing matching outfit colors, happy pose with raised hands. The real person is ultra-realistic photography, natural skin texture, detailed hair, soft smile. Warm golden background with strong bokeh lights, dreamy glow, soft studio lighting, shallow depth of field, premium commercial look, 85mm lens, f/1.8, high resolution, sharp focus on face and mini character, clean composition, realistic shadows and reflections.glm-image result:
you can't just use a sd style prompt and call this model don't work, just stick with sd if you are with that
Just trying to lighten the mood here, if I can I guess. This is a first time I see controversy between 2 leading Chinese AI Image models in hugginface. Eventhough in short sight, it is bad and inappropriate way to respond to engineers and people who worked hard to release this model, but in long sight, this infighting between them will lead to free engagement and American Companies like Google and OpenAI would wish if this controversy was not created in the first place. Just like Oppo, Vivo and Oneplus is fighting to get market share, but they are all part of BBK electronics, if thought from business perspective. From now, others can be even suspicious, if this infighting was manufactured even, which is not to be true. Anyway, there is no thing like bad attention, isn't it?
For previlaged consumers like me that own capable hardware like m4 max macbook pro or others that own nvidia graphics cards, we are really glad to see this infighting between open source AI Image Chinese models. As long they continue to open source it, most of the consumers support them, I am glad they do, or else there might be no difference to support chinese models over American ones, if both are closed ones right. America's Democratic geopolitics is in lot of controversy right now due to trump factor, yet businesses and transparency loving people still trust them compared to their proprietory chinese counterparts.
The issue is that Z-Image is heavily optimized for photorealism, while GLM follows a more general-purpose approach. In short, you’re taking a model designed specifically for realism and comparing it to a general image model, then criticizing it for not producing the same results.
That’s a biased and non-technical analysis, full of gaps, and it’s particularly inappropriate to post it here.
Just trying to lighten the mood here, if I can I guess. This is a first time I see controversy between 2 leading Chinese AI Image models in hugginface. Eventhough in short sight, it is bad and inappropriate way to respond to engineers and people who worked hard to release this model, but in long sight, this infighting between them will lead to free engagement and American Companies like Google and OpenAI would wish if this controversy was not created in the first place. Just like Oppo, Vivo and Oneplus is fighting to get market share, but they are all part of BBK electronics, if thought from business perspective. From now, others can be even suspicious, if this infighting was manufactured even, which is not to be true. Anyway, there is no thing like bad attention, isn't it?
OpenAi and Google are not in this market . They dropped both some open source breadcrum models ages ago (i still like (de-censored) Gemma-3 though), but otherwise they run paid, gated multi bn image/video and estimate 1.8tn language models.
They are not really competing in this private-local users field. Or care about.
On the topic of "first controversy" i also disagree. There is, as in most areas of life, some serious fanboyism going on. Z-image is a favorite there, with people posting everywhere unwanted how much "better it is". Not just here. It maybe, it probably is, but a pointless two random image post like this isn't an argument. Same currently going on between Wan and LTX2, that you need to prompt pamper to kingdom come for a (only then) good result instead of otherwise a unusable mess.
and finally: Z-image got the raised toy arms right but the matching color thing wrong.