Home Creators Posts Import Register Favorites Logout

Content

I've been researching QI-21 a lot lately. Its editing capabilities are truly impressive, and its image generation performance is also quite solid, but there are two unavoidable issues: first, there is excessive noise during sampling, and the image convergence is insufficient. This can be alleviated to some extent via secondary sampling, but that leads to the second problem: the image ends up looking overly greasy. Today's workflow is specifically designed to address this very issue.

Global Comparison

local contrast

Based on QI-21, QI-21-plus optimizes the skin texture, reduces the greasy feeling, fixes the problem of dark images, and meanwhile, its generation speed has been increased by 2 to 3 times.

As usual, this is what the workflow looks like.

Open the subplot, and you will see three regions

Region 1 is the standard single-image editing workflow for QI-21, which loads my LoRA "HighQuality for QI-21" and replaces the Unet with the newly released QI-Turbo. This base model can increase the generation speed by 2 to 3 times without compromising image quality.

You can set the frame size for the initial sampling here. The recommended range is 1024 to 2048.

Connect the image generated in Region 1 to Region 2, which is the optimized single-image editing workflow for Klein. The 9B version of my trained LoRA HighQuality for Klein is loaded here, and the prompts have also been optimized to ensure that richer details are generated while keeping the content of the image unchanged.

Similarly, you can set the size of the downsampled image here.

At this point, you will get a less overprocessed 4K image. If you want a larger image size, you can feed the output generated by Region 2 into Region 3, which contains two sets of upscale nodes: DLSS5 and VOSR2.

Eventually, you will get an 8K high-definition restored image.

Of course, the above parameters are only the default settings for the workflow. If your device has high performance, you can also try using higher parameter values.

The LoRA you may need

✨HighQuality for QI-21✨private version

✨MoreReal for QI-21✨private version

✨Remix for QI-21✨private version

OK! That's all! I hope you like it!

If you enjoy my sharing, you are welcome to join my private club. If you are already a member, you can get the project files for this sharing directly below. 👇👇👇

🫡May the AI-power be with you, see you soon!🫡

Disclaimer: My model or workflow is solely for learning and communication purposes and shall not be used for any commercial activities. All responsibilities for the generated content, including its legality, ethical implications, or potential misuse, are borne entirely by the user. By using this model or workflow, you agree to this principle. Please use it responsibly within the scope of local laws.

pFG9nxWr6jmwGJCtAzRHjsMLLn4arvujg4Wmks1hESU

Files

Comments

No comments found for this post.