setup (once)

• drag pixart-xl-2-1024-ms-q4_k_m.gguf [1GB] to > ./ComfyUI/models/diffusion_models
• drag t5xxl_fp16-q4_0.gguf [2.9GB] to > ./ComfyUI/models/text_encoders
• drag pixart_vae_fp8_e4m3fn.safetensors [83.7MB] to > ./ComfyUI/models/vae

run it straight (no installation needed way)

• run the .bat file in the main directory (assuming you are using the gguf-node pack below)
• drag the workflow json file (below) or the demo picture above to > your browser

workflow

• example workflow for gguf
• example workflow for safetensors

review

• should set the output image size according to the model stated, i.e., 1024x1024 or 512x512
• pixart-xl-2-1024-ms and pixart-sigma-xl-2-1024-ms are recommended (with 1024x1024 size)
• small size model but good quality pictures; and t5 encoder allows you inputting short description or sentence instead of tag(s)
• more quantized versions of t5xxl encoder can be found here
• upgrade your gguf-node (see the last item in reference list below) to the latest version for pixart model support

paper

• pixart-α
• pixart-Σ
• high-resolution image synthesis

reference

• base model from pixart-alpha
• comfyui comfyanonymous
• gguf-node (pypi|repo|pack)