LCM Hyperloop - 8 steps - 14 sec - 12 images版本hyperloop-v37.8 (ID: 228735)

using 4090,

8 steps - 14 sec - 12 images

reduce batch size if you have a lower GPU

I know the workflow is messy

Updated on the workflow of Capybara's SDXL Workflow v1.4

capybara_paws

you will need:

LCM lora - Latent Consistency team

Advanced Enhancer XL LoRA Z_phyr

Add More Details - Detail Enhancer Lykon

models used

as base Cardos XL v1 Charsheetanon

as refiner NightVision XL - Photorealistic 0.7.7.0 Socalguitarist

描述:

images of 1536 x 2688 in 63 seconds!

updates: much better output quality

We added a model merge, which improved the output quality, and now model options are not restricted to Cardos and Nightvision; other models are getting good results as well.

the work flow is now is 10 steps instead of 8, i have seen differency between my GPUs, one does better with 8 steps "4-4" the other does better with 10 "5-5"

2 IP adapters are now IPadapter Plus, and "IPAdapter Plus Face" is separate, having its own node

Added 1 additional "Apply node" for for each IP Adapeter to have more control over the Base model and refiner model IPadapter weight

I forgot to name the "apply nodes for IP Adapters Base and Refiner, The base model control on top, the refiner model control is below.

To improve the speed, change the save image nodes in the work flow to the standard save image node; it will improve at least 16+ seconds for 6 images.

This version's Batch size reduced to 6 images, with an image size of 768x1344, then upscaling to 1536x2688

SDXL Upscaling using RTX4090

(1536x1536 upscale size, not image size) took 63 seconds for 6 images, double the SDXL recommended size

(1024x1024 upscale size, not image size) took 81 secondsfor 6 images, double the SDXL recommended size

12 images in 2 minutes and 1.4 seconds! 1536x2688....

Canva size: 1536x2688

Image size: 768x1344

Scale factor: 2

Upscaling iteration 1 with scale factor 2

Tile size: 1536x1536

Tiles amount: 2

Grid: 2x1

Redraw enabled: True

Seams fix mode: NONE

Requested to load SDXL

Loading 1 new model

100%|████████████████████████████████████████████████████████████████████████████████████| 4/4 [00:16<00:00,  4.11s/it]

Requested to load SDXL

Loading 1 new model

100%|████████████████████████████████████████████████████████████████████████████████████| 4/4 [00:16<00:00,  4.12s/it]

Prompt executed in 121.74 seconds

SD1.5 Upscaling:

I didn't do the speed test as i am still working to improve it

训练词语: