Unable to load image
/h/ai_slop House Vampire
iStillMissEd buter/toast Old strags best strags  8mo ago (text post) Edited 8mo ago 200 thread views #257137

[plz help] SDXL (ponyXL) is giving me some issues.

Been trying to make ponyXL work for the last few hours and I'm not getting anywhere

https://i.rdrama.net/images/1711656964835725.webp

All my outputs look like the above. Simple colours and shapes with no detail. I'm using the SDXL vae (using no vae still doesn't fix it) and not using refiner. Running on a rx580 8gb in medvram mode (otherwise I don't have enough vram to hold the model)

[mostly] Solved! I was using too low of a resolution, bumped it to 1024x1024 and it works better https://i.rdrama.net/images/17116591267189615.webp

13
New Old Top Bottom Controversial Random
Jump in the discussion.

No email address required.

Did you make sure to set the resolution of the output image to 1024x1024? This is a common error (for me at least). SD1.5 etc use 512x512, but XL uses 1024x1024

Jump in the discussion.

No email address required.

I am using 512x512! I'll try 1024 rn

Jump in the discussion.

No email address required.

also remember that ~30 steps should be fine for generation, no need for 150 like in 1.5

Jump in the discussion.

No email address required.

I use LCM, I do 6-10 steps for 1.5 lol.

And it looks like you were right! 1024x1024 nearly crashes my system (Literally can't run anything but automatic1111 and firefox) but it works (at least kinda)!

https://i.rdrama.net/images/1711659082904002.webp

Jump in the discussion.

No email address required.

lol well at least it works haha

Jump in the discussion.

No email address required.

!codecels

Jump in the discussion.

No email address required.

It seems like you're encountering issues with image quality while using ponyXL with the SDXL VAE. The blockiness and lack of detail could be due to the VAE you're using. Here's a step-by-step guide that might help resolve the issue:

Update the VAE:

  • Ensure you have the latest version of the VAE. If not, re-download it and replace the existing one in your models/vae folder1.

Adjust the Workflow:

  • In your workflow, add the “Load VAE” node by right-clicking, selecting ‘Add Node', then ‘Loaders', and finally ‘Load VAE'.

  • Disconnect the “Load Checkpoint - REFINER” VAE from the “VAE Decode” node.

  • Connect the new “Load VAE” node to the “VAE Decode” node1.

Check for Updates:

  • Sometimes, the VAE is updated shortly after release. Verify that you have the updated version by checking the file hash with a command like:

certutil -hashfile sdxl_vae.safetensors MD5

  • The correct MD5 hash should match the one provided in the update notes1.

VAE Settings:

  • If you're still facing issues, consider changing your VAE settings in the settings tab of your software to ensure it's not set to an older version2.

Hardware Considerations:

  • The RX580 8GB might be limiting the performance. If possible, try running the model on a system with more VRAM or in a lower VRAM mode that's compatible with your current setup.

Software Compatibility:

  • Verify that all other software components are compatible with each other and that you're not using any deprecated or conflicting versions.

If you follow these steps and still face issues, it might be helpful to seek assistance from the community where the software is discussed, as they might have specific insights into the problem. Remember to provide them with details about your setup and the steps you've already tried. Good luck! 🛠️

Jump in the discussion.

No email address required.

You had a chance to not be completely worthless, but it looks like you threw it away. At least you're consistent.

Jump in the discussion.

No email address required.

try copying a random civitai gen exactly to see it isn't a prompt issue.

pony is picky about prompts

Jump in the discussion.

No email address required.

try copying a random civitai gen exactly to see it isn't a prompt issue.

That's what I'm doing, copied prompt, seed, size and model

Jump in the discussion.

No email address required.

what do you prompts look like? do you have "score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up," at the beginning of your prompt?

edit: also your amd GPU might be causing problems. i have 16gb vram and i'll run out of memory using ponyXL.

Jump in the discussion.

No email address required.

source_ cartoon, 60s hippie, psychedelic, girl smoking, cannabis plants in background, score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up , xenotrip, monochome, 4k, 8k, realistic

Yea I need to use all my VRAM, all my system ram and 16gb of swap to run it lmao. Turns out I had the res set wrong

Jump in the discussion.

No email address required.

try putting the "score" stuff at the front of the prompt. stable diffusion prioritizes phrases at the beginning of the prompt

good to know you got it to work tho :marseythumbsup:

Jump in the discussion.

No email address required.

"""working""" still crashes my pc half the time. Gonna buy more ram and see if that helps

Jump in the discussion.

No email address required.

you need a nvidia GPU. more ram wont do much.

Jump in the discussion.

No email address required.

:#marseycry:

But they cost sooo much! !peakpoors

Jump in the discussion.

No email address required.

:#marseylgbtflag:

Jump in the discussion.

No email address required.

Link copied to clipboard
Action successful!
Error, please refresh the page and try again.