PixArt Sigma 900M: Enhanced Text-to-Image Model
Model Overview
PixArt Sigma 900M is a text-to-image generation model based on the PixArt Sigma architecture. This version has been expanded to 900M parameters, up from the original 600M base model. Two distinct variants are available:
PixArt Sigma 900M Base (Photorealistic)
PixArt Sigma 900M Anime
Key Features
900M parameters (300M more than the base model)
Improved image generation quality
Technical Details
Architecture: PixArt Sigma variant
Parameter Count: 900 million
Base Model: 600M PixArt Sigma
Credits
Original PixArt Sigma model by PIXART-Σ (pixart-alpha.github.io)
Based on jimmycarter (Jimmy Carter) (huggingface.co) adaptation of PixArt Sigma
ComfyUI Compatibility
To use this model in ComfyUI, a custom node is required as ComfyUI does not natively support this variant at the moment. The necessary custom node can be found at: https://github.com/city96/ComfyUI_ExtraModels
Required Components
T5 Model:
Download from: PixArt-alpha/PixArt-Sigma-XL-2-1024-MS at main (huggingface.co) on Hugging Face
Naming Convention: Ensure the T5 model files are named exactly as they appear in the repository
Important: Remove any prepended "text_encoder" from the filenames if present
Setup Instructions
Download the T5 model from the Hugging Face repository linked above
Rename the T5 model files if necessary, removing any "text_encoder" prefix
Install the custom ComfyUI node from the GitHub repository
Place all model files in the appropriate directories as specified in the ComfyUI_ExtraModels documentation
Description
photorealistic variant
FAQ
Comments (20)
Congratulations, Looking next level!, any chance or plans for You to upload it on tensor art to try it?
thumbs down was a joke because you asked for this to be posted on a competitor. this is a great model I hope you get to use it everywhere and anywhere
@Cosmic_Crafter lol ,oh i never saw it that way, in order to use it, i hope it can happen 😆
All hail Civitai
DV,
I'm getting this and my ape brain is confused. Please help.
"Error no file named pytorch_model.bin, model.safetensors, tf_model.h5, model.ckpt.index or flax_model.msgpack found in directory E:\ComfyUI_windows_portable\ComfyUI\models\t5."
check your t5 models you need to use the one model-00001 of 00002.safetensors the safetensors model is what you need if you have the bin model it is the wrong one
@Cosmic_Crafter Thanks! This was the exact right answer!
could you pls upload your TempestV0.1 here too?
What resolution?
This model defaults to 1024.
Most workflows also use a ratio instead of pixel resolution.
Other PixArt models have other resolutions too!
Results look ok to me, not like wow, but when I try it I get an error: "Error occurred when executing T5v11Loader: Unable to load weights from pytorch checkpoint file for 'C:\Users\JayNL\Documents\ComfyUI_windows_portable\ComfyUI\models\t5\pytorch_model.bin' at 'C:\Users\JayNL\Documents\ComfyUI_windows_portable\ComfyUI\models\t5\pytorch_model.bin'. If you tried to load a PyTorch model from a TF 2.0 checkpoint, please set from_tf=True." and a lot more...
When I set path_type from folder to file it doesn't give an error but the result is a black image, I really have no idea what I am doing 😂 also there is no from_tf=False in any file.
it works on automatic?
tips to let this works in comfyui ?
This vid will show you the way, watch the first 10 mins
https://www.youtube.com/watch?v=9YDR6YuYFtQ&t=543s
@boricuapabPR why it's fucking misery !!? from where i can get the T5 ?? i wont't anymore thanks same like sd3 misery , i don't want to thanks again.
waiting for A1111 version
no a1111..... a1111 is getting unsupported more and more each day... this forcing of comfy ui is kinda pointless. Yes, it's better, but, people just want to generate few images, not spend half of day arranging nodes ;)
Also, it seems to me that this is not-so photo realistic...so... i can wait.
Can u please make an fp16 version? Fp32 crashes comfyui I can't get it to work,but sigma xl 2 works fine
Is this still being worked on?
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.