Downloads · 30 days
136
25% of all-time downloads
Wan2.2-I2V-Weak-ID
rzgar/Wan2.2-I2V-Weak-ID
Wan2.2-I2V-Weak-ID is a text-to-image model from rzgar. Use it when you need an image from a text prompt. It is set up for diffusers. The card lists the license as apache-2.0.
<video controls width="1280" height="720" <source src="https://huggingface.co/rzgar/Wan2.2-I2V-Weak-ID/resolve/main/demo/ComfyUI00005.mp4" type="video/mp4" Your browser does not support the video tag. </video <Gallery /
All-time downloads
539
Public
Repo size
19.7 MB
Likes
2
Public
Repo makeup
Click a slice to open those files.
.mp410.5 MB · 53%
At a glance
- Task
- Text-to-Image
- Library
- diffusers
- License
- apache-2.0
- Access
- Public
- Created
- Jul 19, 2026
- Updated
- Jul 19, 2026
- SHA
- 40eeead7
Base models
Model card
From the Hugging Face model README
Wan2.2 I2V Weak-ID
<video controls width="1280" height="720"> <source src="https://huggingface.co/rzgar/Wan2.2-I2V-Weak-ID/resolve/main/demo/ComfyUI_00005.mp4" type="video/mp4"> Your browser does not support the video tag. </video> <Gallery />Prompt:
- High noise strength: 0.36 | Low: 0.75
(left input image) the young woman with red hair wearing a red high-collared lace-up top, stands in a dim room
Model description
Small hand-tuned LoRAs (3MB) that give basic edit control over a Wan2.2 I2V start image: how hard the reference locks the video versus how much the prompt can reshape the scene..
Higher LoRA strength = weaker identity and stronger prompt influence. (semi Bernini-R image-to-video edit)
Suggested strengths
| Stage | LoRA strength (sweet spot) | Effect |
|---|---|---|
| High noise | 0.25 – 0.36 | More prompt control over the scene (layout, environment, overall look) while still using the start image as a soft guide |
| Low noise only | tune as needed | Tends to shift facial identity while clothing and pose stay closer to the reference |
Useful when you want to borrow style and pose from a reference without fully locking the face—or when you’d rather not run a separate model just to invent a new start image.
LightX2V tips
- Scene changes via prompt with high noise active: prefer Bernini-R LightX2V LoRAs at 2-step (4Step is too strong for Wan2.2 I2V, skin burn, etc,.
- Facial identity tweaks, or low-noise-only runs: regular 4-step LightX2V, or Wan2.2 I2V 2Step (recommended for prompt testing and fast previews).
Download model
Download them in the Files & versions tab.
- Task
- Text-to-Image
- Library
- diffusers
- License
- apache-2.0
- Wavespeed
- live
- Created
- Jul 19, 2026
- Updated
- Jul 19, 2026