Setup · 2:46
Update and model download
Get SD3 Medium.
- Update ComfyUI with Manager → Update All.
- Four SD3 Medium files; skip the one without text encoders.
- Civitai is easiest; Hugging Face asks for details. Low VRAM → 5 GB version.
| Version | Includes |
|---|---|
| sd3_medium | No text encoders |
| incl_clips | CLIP, no T5 |
| incl_clips_t5xxl_fp8 | CLIP + T5 fp8 |
| fp16 T5 | Largest |
Do this in your own tools, step by step. Ticks are saved on this device.
Skip which?
The one without text encoders.
Easiest source?
Civitai.
0:00 welcome to episode 5 of this comfy UI Series today I want to show you how you can run stable diffusion 3 medium on comfy UI where to get the models which model is better and how it compares to the sdxl model first make sure your comfy UI is updated so you can run stable diffusion 3 on it uh click on the manager then click the button that says update all um after the update you will receive a message uh click on restart and then you should have the latest comfy UI version now we need to download a stable diffusion 3 Model the easiest way is to visit the civit AI website then search for stable diffusion 3 it should look like this and is uploaded by this user there are different versions of the model available but which one should you choose if you scroll down you can find a quick start guide by clicking on that link you can read more about it on that page the sd3 media model for example doesn't include text encoders this model includes everything necessary to run including the floating Point 8 version of the text encoder another Model includes clip but does not have the large t5x XL text encoder so if you go back and scroll down to the end
1:12 you'll see that the original model can be also found on hugging face by clicking where it says uh here you can visit that page to see more information about the models and the actual files available for download and keep in mind that to access these models you'll need to provide provide some details such as an email address and you know other information that's why civit AI is the easiest option for downloading we have a total of four different models each in different sizes I won't download the first one because it doesn't include text encoders so you would need to download them separately which I don't really recommend I will download the other three models to test them and and show you the differences if your video card doesn't have much video RAM get the 5 gigabyte version because it includes clip if you have a better video card go for the 10 gigabyte or even the 15 gigabyte version most people can download the 5 gigabyte version because it includes clip click on the download button then find the folder where you want to place the files it should be in the models folder than in um checkpoints just like where you have your other
2:21 stable diffusion models you can see that I have downloaded all three models and these are the three models available on civi I rename them to match the original names so that people using my workflow will know what each one is however you can use any names you prefer for the models if your comfy UI was open make sure to click the refresh button so your models will appear in the interface I've
Get the other 5 chapters, the exam and your certificate for LinkedIn, plus all 700+ AI courses.