Custom Workflow for Z-Image
This workflow works best for my Underview penis model but can easily be changed to what you need. The workflow has pieces of other workflows I've found to work well, with lots of trial and error for getting the best quality using loras without degrading the image. The workflow is managed with subgraphs to make it visually less overwhelming. The custom nodes used in the workflow are below:
rgthree-comfy
ComfyUI-KJNodes
ComfyUI-SeedVR2_VideoUpscaler
RES4LYF
was-node-suite-comfyui
ComfyUI-MagickWand
SeedVarianceEnhancer
comfyui-vrgamedevgirl
v2 of the workflow uses the new RealTimeLora v2 node pack for the Underview v2 lora and the EulerDiscreteScheduler node for the output quality improvement
Description
v2 of the workflow uses the new RealTimeLora v2 node pack for the Underview v2 lora and the EulerDiscreteScheduler node for the output quality improvement
FAQ
Comments (9)
keep getting "No outer link found for slot [2] conditioning" error either ver1 or ver2
What they mean by "outer link for slot [__]" since Comfy decided to be so cryptic without more detail: it's just the left-hand side of nodes with something not connected like VAE, model, positive, negative, latent_image.
Make sure you have wired everything. Models node connect to Prompt node as well as the LoraLoaderModelOnly (Non-Scheduled) node; have LoraLoaderModelOnly (Non-Scheduled) output to LoRA Loader (Scheduled) node and Prompt node connect to LoRA Loader (Scheduled) node. Have the LoRA Loader (Scheduled) outputting it's model_2, positive_2, and negative_2 to all of the Stages in the Image Base section. The image previews for each of the stages are in the middle Compare box and are already connected without wiring. Make sure you have everything installed for node packs that he mentions in the description above or else it won't work at all.
I have no idea what your pc is like, but this took me about 30 minutes to do all of the stages on my RTX 4080 Super, specifically Stage 5 taking most of that time. ChatGPT had to help me for several hours to understand how this workflow worked in the first place since the roadmap is very confusing. The final output was meh....I probably did something incorrect though.....
Also, try right clicking on the subgraph/sampler nodes and then click “Edit Available Widgets”. Sometimes the widget connections don’t transfer and that usually fixes it.
i meet this issue too , have you deal with this problem ? i really need help guys
@q2291268572517 not yet sorry
Very nice, works best when image is shot from down. You are in the right direction compared to other loras. You should definetely train more angles and soft version. This is the best ZIT lora for now. Keep up the good work.
It's hard to put together your workflow. I've been struggling for two days now and still haven't been able to put it together. I keep getting errors. ((((((((
Hello. I downloaded your workflow, but I can't figure out where to include my LORA. No matter where I put it, it doesn't recognize my model's face.
LoraLoaderModelOnly (Non-Scheduled) here or there LoRA Loader (Scheduled)
This is the best lora ever, Plz we need Z image base vision