Stable diffusion thread

[doogyhatts]

it seems local GPUs are quite useless except to generate own prawn and saxy ai images.. hehe

i'm on rtx4060M .. should i upgrade to rtx5070ti desktop? best CP value!

You can wait for the 5080 super or 5070 ti super, both are likely to have 24gb vram.

Local GPU can use Qwen-Image-Edit-Plus to do some image editing.

 

[doogyhatts]

@spunkbrat Are you able to see the post below?

 

[spunkbrat]

@spunkbrat Are you able to see the post below?

Yup, why?

 

[spunkbrat]

My screen shows this rainbow unicorn.

Huh! I see some Orpheus Apothesis tweet

 

[doogyhatts]

Huh! I see some Orpheus Apothesis tweet

I can see the post on X, but not here.
//
Update: I can see the posts here now.

 

[doogyhatts]

 

[doogyhatts]

 

[doogyhatts]

 

[doogyhatts]

 

[doogyhatts]

 

[doogyhatts]

 

[doogyhatts]

https://blog.comfy.org/p/flux2-state-of-the-art-visual-intelligence
https://blogs.nvidia.com/blog/rtx-ai-garage-flux.2-comfyui/

Someone managed to run it with only 12gb vram.
I will wait for Wan2GP to integrate the new model first.

 

[eno_lc]

https://blog.comfy.org/p/flux2-state-of-the-art-visual-intelligence
https://blogs.nvidia.com/blog/rtx-ai-garage-flux.2-comfyui/

Someone managed to run it with only 12gb vram.
I will wait for Wan2GP to integrate the new model first.

Any chance can run on 5080?

 

[AikiBoy]

https://blog.comfy.org/p/flux2-state-of-the-art-visual-intelligence
https://blogs.nvidia.com/blog/rtx-ai-garage-flux.2-comfyui/

Someone managed to run it with only 12gb vram.
I will wait for Wan2GP to integrate the new model first.

No money for good computer.. money need to spend for milk powder lol..

 

[doogyhatts]

Model weights: https://huggingface.co/Comfy-Org/flux2-dev
GGUFs: https://huggingface.co/city96/FLUX.2-dev-gguf

Enhanced Weight Streaming: NVIDIA partnered with ComfyUI to upgrade its "weight streaming" feature, which allows massive models to run on GeForce RTX GPUs by offloading data to system RAM when GPU memory is tight.

Any chance can run on 5080?

Someone on reddit said he was able run it on a 3060 (12gb vram), using the FP8 model.
But he also said it was very slow to generate the image.

I have yet to check the loading time though for GGUF Q_5_KM.
I will wait for Wan2GP to add the model in first before I try it on a 5080 locally.

 

[eno_lc]

Model weights: https://huggingface.co/Comfy-Org/flux2-dev
GGUFs: https://huggingface.co/city96/FLUX.2-dev-gguf



Someone on reddit said he was able run it on a 3060 (12gb vram), using the FP8 model.
But he also said it was very slow to generate the image.

I have yet to check the loading time though for GGUF Q_5_KM.
I will wait for Wan2GP to add the model in first before I try it on a 5080 locally.

The guy seems to mention i2i is really slow

 

[focus1974]

https://blog.comfy.org/p/flux2-state-of-the-art-visual-intelligence
https://blogs.nvidia.com/blog/rtx-ai-garage-flux.2-comfyui/

Someone managed to run it with only 12gb vram.
I will wait for Wan2GP to integrate the new model first.

have to!
majority are buying those 16gb VRAM rtx.
Best CP value!

 

[doogyhatts]

The guy seems to mention i2i is really slow

Hmm... I see.
That's probably because of all the additional processing due to the reference images.

Apparently, there is a significant quality difference between Pro and Dev models.

Some results from using image references.

 

[eno_lc]

Hope the hardware requirements stay the same as 2509

 

[doogyhatts]