How to Use
This is a style LoRA, designed to be the base layer of your LoRA stack. It creates the foundational aesthetic of realism, upon which you can add character or concept LoRAs.
Note: The included ZIP archive contains both the high-noise and low-noise LoRA variants, along with our recommended ComfyUI workflows.
Trigger Word:
InstacamRecommended Strength:
1.0. Start here and adjust in small increments.
I would also like to thank Danrisi, who originally taught us how to train LoRAs and helped make our work possible.
Description
Even better "amateur" look. More low quality images added to the dataset and I took out all the ones with excessive makup, face shinyness. Fully tested and working with character LoRAs, examples shown here.
Alot of the images shown here have a weird dimple thing, but it's more from our character LoRA and not the base LoRA.
FAQ
Comments (22)
Will there also be a flux lora version?
Nope, whole point is that this is far better then flux
Flux is distilled model, and very bad to train we were just using it because there were nothing better, wan now could be a very all in one model, just like Sd1.5 was
If there are further updates, could some young Asian women be included?
Will you be releasing the dataset?
What's the workflow for creating text to image with the video model?
If you set the frame length to 1, WAN automatically uses t2i inference mode
The best was to answer ANY question like this is to use Google (do people not search any more?) with 'reddit' (or huggingface) in the search phrase!
If you click on an image, under "Other Metadata" where it shows x amount of nodes, you can click there to copy and paste it into comfyui
T2V and T2I are the same thing. Video is just a series of images. Set your frame rate to 1.
Looks amazing but I'm struggling to get similar results as the preview images. Would you mind sharing your full workflow?
If you click on an image, under "Other Metadata" where it shows x amount of nodes, you can click there to copy and paste it into comfyui
00quebec Thank you ❤️
i'd also reallly like a workflow!
If you click on an image, under "Other Metadata" where it shows x amount of nodes, you can click there to copy and paste it into comfyui
This looks amazing! Are you willing to share a bit more about the training process? Cosine scheduling, epoch count, batch size, learning rate?
Something i wondered: would this approach also work for img2img? Eg turning a highly "AI" looking image into a more realistic one?
I have not tried that but It seems really intresting! We will look into that.
This workflow says it does img2img with Wan. Might be worth looking at.
How do you daisy chain loras in wan 2.2?
Try the Hunyuan Lora load node
power lora loader and you wont need to daisy chain anything