I've been running it through tests on my local machine. The Text-to-Image component is VERY poor. It's not a good sign when the stock ComfyUI Flux3 Klein template with built-in config/prompts produces SD 1.5 levels of half-melted faces and various other body horror.

Flux 2 Klein 9b also had anatomical issues as well, but this one feels practically lobotomized.

https://imgpb.com/wDGGs

Some examples of the Reddit posts/memes:

https://reddit.com/r/StableDiffusion/comments/1x79gth/why/

https://reddit.com/r/StableDiffusion/comments/1x7ktbu/enshit...

I gave Flux 3 Klein my typical human test that has never failed on newer image models and it did fail; in addition to lower quality, the fingers were definitely not the correct proportions.

Whether the cause is due to more safeguards against NSFW content is hard to confirm technically; it was more an intuition back in the Stable Diffusion XL days but I suspect modern training paradigms are better at correlative inference.

The model is perfectly safe if nobody wants to use it!

Yes, a prime example of when the performance vs "bias" tradeoff is handled anti-business

Didn't know about that but it makes sense

Was creating characters for a clients social media avatar and so was doing various tests with the latest bunch of models and omg the characters from flux were like deliberately ugly I was shocked

Lame. Back to Chroma it is then.

There's a base version for LoRA training so a porn adapter is likely already being built.

I've tried it some, only locally. I rarely use APIs, so can't say what the official inference API results look like. It correctly followed some prompts that almost every model I've ever tried performed poorly at, but failed in strange ways on prompts that other models handle with ease.

At high resolutions, structural integrity on people can definitely fall apart like older models, where you get one person turning into a long spaghetti spanning across the frame. It might be that you need to do grounding by generating some kind of scene reference at its stable resolutions first then having it use that for the larger resolution like older models too.

They say that their prompt "upscaler" model or system prompt are essentially necessary for the model, but my results have been hit or miss with that claim. Some basic prompts are better, some upscaled prompts are better. I think that if you gain an intuition for how the model understands things, you might get better results by default by hand crafting prompts specifically for it.

Some of the stylized art I'm getting out of it at higher resolutions are really nice and I prefer them over other models.

All of these models seem to have their own character in terms of system requirements, performance, prompt adherence, artistic outputs and so on. None of the models I have tried are a complete homerun, so maybe in this case the model feels a little inconsistent in some ways, but it's a model I will probably keep around.

The complaint has some basis but is greatly exaggerated.

When the whole stack comprising the model is used, it occasionally produce erroneous anatomy and require running qitha different seed to get a good result; this is much less common than with Flux 2 Klein. Also, the step distilled version of the model, particularly, isn't great at generating photo-quality skin, despite otherwise being good at producing photographic-seeming outputs. (Even some of that might be exaggerated, because it was what I saw before fixes to the Comfy templates.)

But a lot of people on Reddit (and some here, I would guess—none are explicit about it but that's the only thing I can think if that would make the “worse than Flux 2” comments remotelt understandable) have decided that they don’t like that BFL’s approach to generality in prompt handling for the model was to train it with a particular style of detailed prompt and provide a LLM-based prompt rewriter (they call it the “prompt upsampler”) to process raw user prompts into the style of prompts the model accepts, and despite BFL flat out saying the model should bot be used without either the provided upsampler or some alternative one (e.g., the published system prompt used with a different LLM), and that that will produce poor results, have decided that the correct way to judge the model is by not upsampling prompts, and pretending that the upsampler isn't part of the stack. This makes generations far worse (it especially increases body horror, probably—from trying it and watching latent previews with and without—because it makes prompts more ambiguous making it harder for the diffusion process to resolve where new elements should be drawn or extra drafted elements should be pruned early in the generation process.

The problems being complained about aren't mostly what you'd expect from NSFW avoidance; it does have some of that, too, even though it is not particularly shy (or low quality) about nudity so long as the genitals specifically aren’t visible. It’s not great with women’s nipples/areolae, but it doesn't avoid them. Its clearly undertrained on NSFW material, but it doesn't look particularly moreso than most corporate base models; its usual for community LoRAs and finetunes to close that gap.

Bounding box support is nice, though the way it is aonewhat soft guidance makes it tricky to use reliably for straight t2i (it looks like it may be atronger for tying references tobit) and, while the Comfy workflow has a tool to help draft it, I think there are some wrinkles with upsampler interaction with it that the workflow doesn't account for.

That said, and despite being great and creative as a first pass composition model at high res that I will definitely keep in my t2i toolbox, it is definitely generally underwhelming for straight t2i, especially given that MOSS Image 5B just dropped, which despite its claim that its strength is in text generation and structured layout (a popular focus for recent models), is one of the best all-round base models I’ve seen in breadth of content knowledge, prompting flexibility, output quality in its strongest resolution range (seems to be 1-2MP for general purpose—full-frame “artistic”, not its structured layout text heavy focus—work), creativity, lack of censorship or undertraining, etc.

From what I saw Flux 2 Klein or Qwen Image 2.1 might give you a less frustrating experience at comparable or better performance. We shouldn't be dealing with body horrors in 2026. Feels like a major regression.

You could argue that denying and perverting our core humanity is what we're all about in the current age...

Looks like the people over /r/SD seem to like MOSS (i'm not up to date on the image stuff) but what i notice is that it is Apache2 licensed when every other openweight model seems to have some weird license (from a quick check Qwen-Image even downright forbids any commercial use at all).

The smaller we go with good models, the more I wonder about tiny animals.

By memorizing the fact that the sun rises in the East and sets in the West, you no longer have to make complex predictions and draw arcs every day. By memorizing the fact that the Earth rotates, and that generally most bodies in solar systems do, you can extrapolate this fact to any planet or solar system.

This example is a rough comparison to what actually happens, but it does attempt to demonstrate what compression and knowledge are, and how patterns, symmetries, and extrapolation/interpolation can make describing/modeling a complex system much more efficient.

Models can even go a step further, offloading entropy to the input. An efficient model exploits predictable structure; the remaining details can be expensive to compress further. The input can supply whatever additional information is necessary to produce the desired output within some margin of error. The information required depends on the model’s efficiency over a given domain, the structure of that domain, and the required precision of the output.

Consider generating an image of a pebble driveway. Most models today will generate a plausible arrangement of pebbles using their learned distributions. Reproducing a particular arrangement requires the input to convey the details that distinguish it from other plausible arrangements. The more closely you want to reproduce it, the more informative the input needs to be—potentially including a reference image of the arrangement itself.

This certainly raises a lot of open questions that are fun to pursue, such as the smallest possible model that can recover a complete description of our universe from minimal input.

That's true. What's also true is that it cannot compete performance-wise.

There is a story that Stability.AI went under with their trash release of Stable Diffusion 3, and that their better people went on to form and work at Black Forest Labs… apparently “3” is viral infected for these people.

this is the right answer. ultimately text and general llm usage incl coding is far more internally complex than just memorizing the looks and composition of concepts placed in an image. dont forget that most of our reasoning benchmarks basically require the llm to have a search engine and calculator in its own weights.

> since an LLM should be able to describer [sic] an image losslessly

Any lossless text description of a given image would just be a structured text output of the exact pixel values, or an encoding that losslessly exploits regularities. Anything less is categorically lossy. An ordinary semantic description of a reasonably complex image will generally be lossy. A short, lossless text description would need to exploit regularities in the image, such as a solid-color image, or a basic pattern or sequence which is easily reproducible, such as a precisely specified checkerboard image.

And being able to represent images losslessly in text simply establishes text’s capacity to encode information. It doesn't establish an LLM’s holistic understanding of images. There is heavy lifting in the phrase “images are a subset.” An image can be almost any conceivable perceptual arrangement of data. The bounds are not clear. Maybe what you meant to suggest was that images of our natural world contain exploitable regularities.

> Natural evolution has also had a much easier time coming up with vision models (everything with eyeballs) than complex language models (as far as we know only a smallish family of monkeys).

It isn't appropriate to explain complex biological systems in terms of today's technologies. This is something people have been doing for a long time. The mind is like a a book, a loom, a camera, an engine, a computer, an LLM/vision model. Etc. The cognitive abilities required for an organism to meaningfully interpret its vision can possibly largely be the same cognitive abilities underlying speech, and all that is missing for a brain to exploit this might be sufficiently advanced vocal cords or manual dexterity; nothing to do with the brain itself, but its auxiliary organs.

When it comes to reasoning specifically, it's a far simpler explanation IMO. Modern RLVR techniques like GRPO rely on the model understanding the likelihood of the generated phrase. This is really easy to compute for LLMs that are trained with next-word prediction. Diffusion models have a harder time accessing this likelihood value (i.e., you can approximate it but the approximation isn't good). Hence, reasoning doesn't work (yet) for diffusion models. Though, alternative approaches are currently being tried out in academia; so someone might crack it!

When you say "they", what are you referring to? Image generation models? LLMs? All of them or older/more recent ones?

When you say "look the same", what do you mean?

I understand this was just a casual one-off comment, but as presented I'm having a hard time understanding what you're trying to convey and if you have appropriate domain knowledge: Your "looks the same" and my "looks the same" might be vastly different, especially depending on context.