rzgar/Wan2.2_I2V_R1_base_v2

This is an improved version of wan2.2-i2v-A14B-Uncensored-base. The base model has been changed to KJ FP8, and an improved lightx2v (tuned specifically for this version) has been merged into it. This version not only works great with I2V but also performs very well in text-to-video style. Use a blank image or the canvas.png provided in the demo folder.

As a rule of thumb, and to preserve the flexibility of the model, the LoRAs are merged at a strength of the “sweet spot” - 20%. The model responds very well when used with style LoRAs or any other wan2.2 I2V and T2V LoRAs, without overpowering the base model. If you use your own LoRA stacks with this model, try strengths in the range of 0.5–0.65. The model already understands the majority of explicit prompts and the mechanics behind them. Using NSFW-Wan-UMT5-XXL-V2 will improve it further.

• (Optional VAE): Cinematic The Matrix theme | B & W

Image To Video Clips

Text To Video style. Input image: canvas.png

Files

File	download
Wan2.2_I2V_High_R1_n54w_v2.safetensors	High noise
Wan2.2_I2V_Low_R1_n54w_v2.safetensors	Low noise

Appearance-Only Enhancers

Rarely needed in I2V scenes, but in T2V scenes they can enhance appearances if the visual results otherwise look underfit.

File	Download	Strength
male_genitalia_enhancer_high.safetensors	High noise	0.55
male_genitalia_enhancer_low.safetensors	Low noise	0.55
female_genitalia_enhancer_high.safetensors	High noise	0.5
female_genitalia_enhancer_low.safetensors	Low noise	0.5

---