My 2-stage (base + refiner) workflows for SDXL 1.0 in ComfyUI, with separate prompts for text encoders. If you don't need LoRA support, separate seeds, CLIP controls, or hires fix - you can just grab basic v1.0 workflow. Works with bare ComfyUI (no custom nodes needed).
SD+XL workflows are variants that can use previous generations SD models to provide initial scene concept.
Links to SDXL 1.0 model files:
sd_xl_base_1.0.safetensors or sd_xl_base_1.0_0.9vae.safetensors* (6.94 GB),
sd_xl_refiner_1.0.safetensors or sd_xl_refiner_1.0_0.9vae.safetensors* (6.08 GB),
sd_xl_offset_example-lora_1.0.safetensors (49.6 MB).
* versions with fixed issue #4310.
Link to ControlNet 1.1 model files (used in SD+XL v1.2):
Link to my SD 1.5 variant used in SD+XL workflow:
MoonRide Mix 10 (you can replace it with any other SD variant you like).
Parameters to play with:
text in prompts,
width and height (but stick to combinations used during SDXL training - listed in notes section),
noise seeds,
steps ("end_at_step" is switching point between base and refiner - should be about 50% to 95% of total steps, depending on selected scheduler and samplers - look at preview image from base to check what is being passed to refiner stage),
scheduler,
sampler names,
cfg (lower - more creative, but also less defined),
CLIP Skip (-1 will treat prompts more literally, -3 will generalise),
LoRA strengths (0 means disabled).
Cheers, and have fun!
Description
Changes since v1.1:
added CLIP Skip controls.
FAQ
Comments (5)
Nice, but where can i find the lora sd_xl_offset_example-lora_1.0.safetensors
You can find it on the official HF page: https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/blob/main/sd_xl_offset_example-lora_1.0.safetensors
@ArtificeAI thx
@ArtificeAI What does it do?
@zGenMedia Basically it adds contrast to outputs, so you get better dark or bright scenes. The official description for it is:
When applied, it will extend the image's contrast (range of brightness to darkness), which is particularly popular for producing very dark or nighttime images. At low percentages, it improves contrast and perceived image quality; at higher percentages it can be a powerful tool to produce perfect inky black.
This small file (50 megabytes) demonstrates the power of LoRA on SDXL and produces a clear visual upgrade to the base model without needing to replace the 6.5 gigabyte full model. The LoRA was heavily trained with the keyword contrasts, which can be used alter the high-contrast effect of offset noise.
Recommended strength: 50% (0.5). The keyword contrasts may alter the effect.