Workflow Ep51 Cosmos Predict2 14b Q8 Gguf Text2image

• status
• prompt_id
• output_images

Model Requirements

• vae: wan_2.1_vae.safetensors -> models/vae
• clip: oldt5_xxl_fp8_e4m3fn_scaled.safetensors -> models/clip
• diffusion_model: cosmos-predict2-14b-t2i-ex3-q8_0.gguf -> models/diffusion_models

Custom Node Requirements

• comfyui-gguf

Links

• https://discord.com/invite/gggpkVgBf3
• https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/vae/wan_2.1_vae.safetensors
• https://huggingface.co/calcuis/cosmos-predict2-gguf/resolve/main/cosmos-predict2-14b-t2i-ex3-q4_0.gguf?download=true
• https://huggingface.co/calcuis/cosmos-predict2-gguf/resolve/main/cosmos-predict2-14b-t2i-ex3-q8_0.gguf?download=true
• https://huggingface.co/calcuis/cosmos-predict2-gguf/tree/main
• https://huggingface.co/comfyanonymous/cosmos_1.0_text_encoder_and_VAE_ComfyUI/resolve/main/text_encoders/oldt5_xxl_fp8_e4m3fn_scaled.safetensors
• https://www.youtube.com/@pixaroma

Routing Metadata

• Family: txt2img
• Input modalities: text_prompt
• Output modalities: image/png
• Model families: sd3, wan
• Node count: 10
• Complexity score: 6
• Resource profile: high
• Estimated runtime: slow (often 2-6 min depending on model/server load)
• Max latent resolution hint: 1024x1024
• Max sampler steps hint: 35

Detected Models

• vae: wan_2.1_vae.safetensors -> models/vae
• clip: oldt5_xxl_fp8_e4m3fn_scaled.safetensors -> models/clip
• diffusion_model: cosmos-predict2-14b-t2i-ex3-q8_0.gguf -> models/diffusion_models

Detected Custom Nodes

• comfyui-gguf

Runtime Warnings

• Large model(s) detected; ensure enough VRAM and disk space.
• Medium-high resolution detected; expect moderate extra runtime.
• Uses custom nodes; missing nodes can cause validation/runtime failures.