If you appreciate the work I have done, I kindly request that you take a moment to provide your feedback. Your input holds significant value for me and will contribute to the enhancement of my future endeavors.
Available on Tensor.art with Free GPU acceleration
Recommended parameters for both versions are almost the same as before, with a few improvements and changes.
Achieving super high-quality images with super low steps and CFG Scale is actually really challenging. I performed countless experiments, and each time, I didn’t achieve what i wanted.
So, I decided to improve the previous version. Now, this version is pretty good. With in 8-10 steps and 4-5 CFG Scale, you will get results comparable to the normal SDXL with 30-40 steps.
✅ Recommended Parameters ✅
Refiner: Not required for this model.
VAE: Use the SDXL VAE.
Sampling Method: DPM++ 2M Karras is effective with 8 steps or more.
Sampling Steps: Start with 8, can increase to 16. Safe starting point is below 7/10. For Euler Sampling Method, 4/5 steps work well.
CFG Scale: Optimal range is 3 to 6. A setting of 4/5 has proven effective.
Hires Steps: Recommended between 4 and 10.
Denoising Strength: Ideal range is 0.1 to 0.5.
Check out more of my exclusive models!
Join my community, Share your feedback, learn, and have fun with us! 😊
Description
Please Note: As this model is still under development, you may encounter occasional imperfections or inconsistencies in the generated images.
Feedback and Contribution: We welcome your feedback and suggestions as we strive to refine this model. Please share your experiences and provide any insights that could help us enhance its capabilities.
FAQ
Comments (2)
Thanks for posting the model. Compared to other Turbo models, yes, the CFG should be higher. For my one run, I like the CFG of 6 best, but now I'm going to do more of a random run. I posted an X/Y plot of CFG vs. steps. Cheers.
No doubt, the best SDXL Turbo so far.
carry on bro🤗
Looks like we don't have an active mirror for this file right now.
CivArchive is a community-maintained index — we catalog mirrors that volunteers upload to HuggingFace, torrents, and other public hosts. Looks like no one has uploaded a copy of this file yet.
Some files do get recovered over time through contributions. If you're looking for this one, feel free to ask in Discord, or help preserve it if you have a copy.