If you like what I do, leave me a ❤️ to receive notifications of future updates and follow me here on Civitai. Remember that it is a hobby for me and it costs a lot of time and everything is shared entirely for free, I appreciate that you make a review and share your creations on this wall.<hr /><h3 id="project-zeus-sd3-v1-vkblwi37w">Project Zeus SD3 V1</h3>After many weeks of training and testing a lot of things, I can finally release this V1 version, and there is good news and bad news.I must say that SD3 is a very capable model, and what it does well, it does very well, it even does some incredible things that sometimes amaze me.The serious problem is that it is a model that has been released broken, rushed and half-finished and what it does badly, it does too badly and what is worse, it has no solution.It is impossible to fix the anatomy, the hands, the limbs of animals in general, not with training, not with LoRa, not with embedding, not with anything. The model is completely broken and bugged. It is simply half-done.I hope that Stability AI gets back on track and really releases a version 3.1, but at this point, I personally highly doubt it. As the model is, it would be like making a practically new model again. But that will be seen, we will always have Flux1.It's a real shame, because SD3 makes incredible landscapes, very detailed portraits, recreates styles very well (the anime and cartoon are especially very effective), but if you try to make a fucking decent hand at it, you can spend the whole day without achieving it.Well, nothing more to add, I hope that this v1, which will obviously be the last, is to your liking and that you can get something positive out of it. A hug to everyone and see you in the next updates of my Flux1 models, <a target="_blank" rel="ugc" href="https://civarchive.com/models/720719/project-gaia-flux1d-uncensored">project Gaia</a> (Dev) and <a target="_blank" rel="ugc" href="https://civarchive.com/models/766618/project-hades-fast-quality-flux">project HaDeS</a> (Schnell).P.D. It's a no clip verson.<h3 id="project-zeus-sd3-beta-x-b3ewkkcef">Project Zeus SD3 Beta X</h3>Small and final fixes before the next version is released. This will be the last Beta version.I am currently working only on training hands and human anatomy in general from scratch. I have had several failed trainings, but at least it has helped me to better learn what modifies what. The next update will take a few weeks to arrive, due to lack of time.This Beta X version is very valid for some things as shown in the sample images that do not have any upscale. After trying Flux, I must say that I am even more committed to bringing this model forward. Flux is very good and does incredible things light years away from Stable Diffusion 3. But my PC burns in hell generating images with it, and it takes too long for me, at least for now. SD3 in that sense is much more manageable for me, and I suppose for many more people like me with a modest PC.We'll see where this journey takes us...I hope you'll join me on the journey of Project Zeus SD3.See you in the next update.<h3 id="project-zeus-sd3-beta-v0.6-0j9ahh9bv">Project Zeus SD3 Beta v0.6</h3>We're out of the Alpha testing phase and I finally have a finished Beta version of Project Zeus SD3 to publish and show you all.What are the differences between the Alpha and Beta versions, you might ask?Well, the answer is easy and not so easy. The font change is actually more for me than for the user who downloads the checkpoint and starts generating images. I consider that I have already acquired some knowledge about the Stable Diffusion 3 model and I'm not going so blind anymore.There is so little information about SD3 training that the concept of trial and error is constant. And it takes a lot of hours because the few tools that exist for it are not well optimized.Despite the current bad reputation of SD3, I must say that I am pleasantly surprised by the potential it has, it is much more powerful than I expected, despite the fact that the release version of the medium model is a real piece of junk. But there is hope, I am convinced that in a few months there will be a checkpoint that will be on par with the best of SDXL. But we'll see about that, the license, despite having changed, is still a big problem at a commercial level.Meanwhile, I continue to enjoy racking my brains and publishing this first update of Project Zeus SD3.I hope you enjoy it.Sampling method: Euler Simple or DPM++ 2M SGM UniformSteps: Between 25-35CFG Scale: 4-6P.D. You need to download the necessary clips.<h3 id="project-zeus-sd3-alpha-v0.5-test-l681c272a">Project Zeus SD3 Alpha v0.5 TesT</h3>This checkpoint is the result of my preliminary tests and trials with the new Stable Diffusion 3. I have no doubt that despite its hasty release, SD3 has a great future. As well as modders, there is a lot of work ahead.Here is the result of my first two weeks of work, I hope you enjoy it. It is an Alpha version, it has millions of errors, but you can get some pretty decent stuff out of it.P.D. You need to download the necessary clips

Project Zeus SD3

projectZeusSD3_v10.safetensors

v1.0

Beta X

Beta v0.6

Alpha v0.5 TesT

If you like what I do, leave me a ❤️ to receive notifications of future updates and follow me here on Civitai. Remember that it is a hobby for me and it costs a lot of time and everything is shared entirely for free, I appreciate that you make a review and share your creations on this wall.<hr><h3 id="project-zeus-sd3-alpha-0.5-test-wzowwxezy">Project Zeus SD3 Alpha 0.5 TesT</h3>This checkpoint is the result of my preliminary tests and trials with the new Stable Diffusion 3. I have no doubt that despite its hasty release, SD3 has a great future. As well as modders, there is a lot of work ahead.Here is the result of my first two weeks of work, I hope you enjoy it. It is an Alpha version, it has millions of errors, but you can get some pretty decent stuff out of it.P.D. You need to download the necessary clips

Project Zeus SD3 V1

After many weeks of training and testing a lot of things, I can finally release this V1 version, and there is good news and bad news.

I must say that SD3 is a very capable model, and what it does well, it does very well, it even does some incredible things that sometimes amaze me.

The serious problem is that it is a model that has been released broken, rushed and half-finished and what it does badly, it does too badly and what is worse, it has no solution.

It is impossible to fix the anatomy, the hands, the limbs of animals in general, not with training, not with LoRa, not with embedding, not with anything. The model is completely broken and bugged. It is simply half-done.

I hope that Stability AI gets back on track and really releases a version 3.1, but at this point, I personally highly doubt it. As the model is, it would be like making a practically new model again. But that will be seen, we will always have Flux1.

It's a real shame, because SD3 makes incredible landscapes, very detailed portraits, recreates styles very well (the anime and cartoon are especially very effective), but if you try to make a fucking decent hand at it, you can spend the whole day without achieving it.

Well, nothing more to add, I hope that this v1, which will obviously be the last, is to your liking and that you can get something positive out of it. A hug to everyone and see you in the next updates of my Flux1 models, project Gaia (Dev) and project HaDeS (Schnell).

P.D. It's a no clip verson.

Files with SHA256: 69823bfb0dc1dec05e131845292a0dd8fc50b977651e850dd9d3878234edce2a