Nothing new, just my spaghetti sauce for a ~30-second video. Maybe it'll be useful to someone.

Control of prompt, loras for each step ( 5 steps ) 
Optional : ollama vision replace with local model if you have enough VRAM 

--

Models:

https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/blob/main/split_files/diffusion_models/wan2.2_i2v_high_noise_14B_fp8_scaled.safetensors

https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/blob/main/split_files/diffusion_models/wan2.2_i2v_low_noise_14B_fp8_scaled.safetensors

Text encode

https://huggingface.co/NSFW-API/NSFW-Wan-UMT5-XXL/blob/main/nsfw_wan_umt5-xxl_bf16.safetensors

Clip Vision

https://huggingface.co/ricecake/wan21NSFWClipVisionH_v10/blob/main/wan21NSFWClipVisionH_v10.safetensors

Lora

https://huggingface.co/lightx2v/Wan2.1-I2V-14B-480P-StepDistill-CfgDistill-Lightx2v/blob/main/loras/Wan21_I2V_14B_lightx2v_cfg_step_distill_lora_rank64.safetensors