JonXL's HQ Photo Creator (2.1 / 2.2 Low Only)
Try clicking the preview images to see the high resolution and detail.
Create High Quality Images with JonXL's Wan Photo Creator.
Updated to V2! V2 unlocks the optional Bleach Bypass Movie FX effect, very useful for creating movie style imagery. (see examples, no lora used)
If for some reason download links are unavailable, please visit https://huggingface.co/JonXL to find a copy of the relevant files, use at your own risk)
Uses only the Low Model of Wan 2.2, meaning you can freely use your Wan 2.1 and Wan 2.2 Low Loras without issue. No need for Wan2.1, High model or High Loras, it just works :)
Note: Also contains a version with Kijai's Wrapper for experts, which seems to produce better results than default K-Sampler.
If you can, please use the expert workflow with a Q5 or Q8 model for best results.
Download links of the required models and instructions are provided in the workflow.
Features:
Supports both K-Sampler based sampling as well as Kijai's Wrapper (including blockswap and offloading)
Supports both 2.1 and 2.2 Low Loras + multi Lora loader
Fast and Quality preset workflows included
GGUF and Safetensors support (toggle/dropdown)
Lightning Lora support (toggle)
Post Film Grain effect (toggle)
Uses RES_2s + Bong Tangent (K-Sampler) & DPM2SDE/Beta (Kijai) for resolutions up to 1920x1080 and 1440x1440
Optional upscale with ESRGAN_2X or other model) (toggle)
Single and multi image generation (+ optional Sage Attention)
Download links and instructions included
Tip: When using 2.1 lora on 2.2 model, it can help to increase lora strength to 1.1 or a bit higher for better results.
Leave a comment, like and post showcase SFW pictures at the gallery below, if you want! Happy prompting! - JonXL
Please do not redistribute this workflow in any way, ask for approval first. For sharing please use a link to this CivitAI page.
Description
Added Bleach Bypass Effect
FAQ
Comments (2)
I find all high quality image generation workflows don't use the high model, which makes them unusable with loras that depend on it for setting up the scene. Would love a lora-capable workflow if you think you are up to the challenge.
Hey Clocksmith. Thanks for your feedback!
There is a reasoning behind why the high quality image generation workflows don't use the high model. It's because it was meant for pre-noising, and from my experimentation, mostly used in driving motion for videos. Since images are not video, there is no need for the high model, in fact it seems better to not use it at all.
For the best results, using this workflow, you would only need to use the low lora from whatever project you downloaded or trained from. I doubt anyone only trained a high model version of their lora, unless it was for driving video, but even then they would have a low version to clean up the noise. A better question would be why lora trainers didn't use their scene dataset images in the low model, since it has to clear up the detail anyway.
Regardless, I already have a workflow for T2V that you could use with both high and low loras, if you really need it for some reason. Set the num_frames to 1. Since it's an older workflow it still uses the K-Sampler which is not the best however, though with res2s + bong tangent it's not bad.
You can find it here, it's GGUF based:
https://civitai.com/models/1862532/wan-22-t2v-gguf-4-steps-upscale-interpolate-clear-vram
It's really not an issue of not being able to add a second sampler (I did it countless of times myself), it's just that for image generation in my observation there is no place for it :)
Hope this helps!