rzgar/FastVideo-FastH3-Trim-Motion-Enhanced-Comfy

Use the companion LoRA with other quantized models released by FastVideo. Strength 1.0 (extracted deltas from v2).

File	Link
fastvideo_fasth3_trim_8step_companion_LoRA_Comfy.safetensors	Download

V2 : About 8% better visuals (8-steps only)

---

V1: Performs well in I2V scenarios, but visuals suffer in T2V scenarios, especially in full-body shots and at a distance

First Clip: 8-Steps, Second Clip: 4-Steps

8-Steps:

• Sampler: er_sde
• Scheduler: beta

4-Steps (v1):

• Sampler: heun
• Scheduler: sgm_uniform

What's different from the base

• more accurate male and female anatomy
• a better understanding of NSFW-related motion (wan2.2 prompting style)

File	Link
fastvideo_fasth3_trim_8step_bf16_motion_enhanced_v1.safetensors	Download
fastvideo_fasth3_trim_8step_bf16_motion_enhanced_v2.safetensors	Download