StableDiffusion version 3 released! First tests
As you may know, Stable Diffusion 3 has been released.
To test this text to image model (txt2img), you will have to go to the huggingface page of StabilityAi dedicated to download Stable Diffusion version 3.
To download the model, you have to sign up, but don't worry it is free of charge if you don't use the model for a commercial purpose.
If you want to use it, you will have to use ComfyUI (at least for the moment).
As I explain in this video, SD3 won't work with Automatic1111 for the moment,
First of all, update or install Comfyui on your computer (if it is not done yet).
I am not going to explain that here.
Then you need the files of a StableDiffusion 3 model:
Put at least one of these files:
sd3_medium.safetensors
sd3_medium_incl_clips.safetensors
sd3_medium_incl_clips_t5xxlfp16.safetensors
sd3_medium_incl_clips_t5xxlfp8.safetensors
Into the ...\ComfyUI\models\checkpoints folder
You also need to put the files from this URL :
clip_g.safetensors
clip_l.safetensors
t5xxl_fp16.safetensors
t5xxl_fp8_e4m3fn.safetensors
Into the ...\ComfyUI\models\clip
folder.
Then use the workflows given here:
Here is a screenshot of the settings that I used with an upscaling workflow:
comfy_example_workflows_sd3_medium_example_workflow_upscaling
The first time you run the workflow, it is very slow...
Do you want to see some outputs ?
Well, here is one:
With the default prompt of the upscaler:
And this is with the basic workflow with this prompt:
A cute and funny cat holding a sign with "m14w.com" written on it. The overall style of the artwork is reminiscent of fantasy or supernatural genres
I am not so impressed: the model is not very fast and the text output is not always consistent with the prompt.
With the multi-prompt workflow (other seed)
A last comment: i find the model pretty a bit slow and the models need time to load, but it could be worse...