setup (once)

• drag hyvid_lora_adapter.safetensors [323MB] to > ./ComfyUI/models/loras
• drag hunyuan-video-t2v-720p-q4_0.gguf [7.74GB] to > ./ComfyUI/models/diffusion_models
• drag llava_llama3-q4_0.gguf [4.68GB] to > ./ComfyUI/models/text_encoders
• drag clip_l_fp8_e4m3fn.safetensors [123MB] to > ./ComfyUI/models/text_encoders
• drag hunyuan_video_vae_fp8_e4m3fn.safetensors [247MB] 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 demo clip or the workflow json file (below) to > your browser

workflows

• example workflow for gguf (upgrade your node for llava gguf support; more quantized llava-llama3 version here)
• example workflow for safetensors (fp8 scaled version [13.2GB] is recommended)

review

• more stable output if adapter applied
• working good even with fp8_e4m3fn scaled clip and vae
• significant changes in loading speed while using the new quantized/scaled file(s) with revised workflow (see above)

references

• lora adapter from trojblue
• base model from tencent
• fast model from fastvideo
• comfyui from comfyanonymous
• comfyui-gguf city96
• gguf-comfy pack
• gguf-node (pypi|repo|pack)

appendices

• get more test prompt from the training dataset
• still no ideas? get fresh prompt from our simulator