This is a Wan2.2 Bernini-R video workflow that supports:

1. Local Gemma 4 gguf file for LLM prompt enhancing.

2. Input of both images and videos for LLM.

Example use case: Video editing with reference images.

Custom node used:

ComfyUI_Simple_Qwen3-VL-gguf

• Set "cpu_moe": true to speed up LLM with limited vram. (Q8 model needs ~6GB vram.)

• To set image_min_tokens for Gemma 4, follow n_ubatch > image_max_tokens > image_min_tokens. (e.g. 2240, 2240, 560)

• Qwen 3.6 can also be used instead of Gemma 4. Just set image_min_tokens to 1024 and n_ctx to a larger value (e.g. 16384).