If you like my content, support me at:
https://ko-fi.com/duchaiten
Patreon is here (but there's nothing in it):
https://linktr.ee/Duc_Haiten
I created my Discord server. I'm still quite new to Discord server management, feel free to share your opinions.
https://discord.gg/vKEW6jqa49
PLS Read this article to learn how to use the Snapvu model.
My wish is to create a model with detail down to every pore on the face.
Don't use hires fix on close-up images, it reduces skin detail.
The settings in the image are for the version without Snapvu.
Negative prompt:
(worst quality:1.5), (low quality:1.5), (normal quality:1.5), lowres, bad anatomy, bad hands, multiple eyebrow, cropped, extra limb, missing limbs, deformed hands, long neck, long body, signature, username, artist name, conjoined fingers, deformed fingers, ugly eyes, imperfect eyes, skewed eyes, unnatural face, unnatural body, error, painting by bad-artist
Description
FAQ
Comments (18)
👍🏻👍🏻👍🏻
read your article. keep up the good work, I will test this one with impatience. I am not yet sold on Noob Ai feels, but I know you can pull things up. I will try to post some nice pics.
I get only noise from the example portrait workflow, even without changing anything ?? ??
read this
https://civitai.com/articles/14482
The snapvu version is an amazing attempt, but so far I've found some problems, for example when I use some of the prompt words that are commonly used in NoobAI, no matter how much I tweak the generation parameters (including the progressively lower cfg), I get images with too high cfg.
But when I use natural language, such as the cue words in your example, the effect of seed on the image becomes smaller, or when focusing on the character, the change in seed does not lead to excessive changes in the character's facial features, even if it removes those details specified in your cue words. This is not the case with your previous (non-noobai) model. The model I used as a comparison was KandooAI's just released ragnarok_by_RunDiffusion.
send me the image you have problem with that image data
@DucHaiten It's been posted below, probably still under review, not sure you'll see it
I have a question: are the NoobAI models compatible with the loras pony, illustrious and sdxl?
I think it is possible, but the effects will be a bit different and also depends on how lora is trained.
@DucHaiten Thanks, but what are the differences between NoobAI and illustrious or pony? is it better? not just your model but generally.
@Suzanne Noobai pony illustrious, are all models retrained from scratch with SDXL architecture, you can simply understand it as houses being demolished and rebuilt on the same piece of land. The finetune or dreambooth models are just expansions or interior and exterior decorations for the model that has been completed. Noobai pony illustrious, different in the huge data they train, how they train, training method...etc
the most obvious difference is that Pony aims to divide quality based on scores to highlight the training data more, with the least amount of training data. illustrious has more training data than pony, the generated images have the largest resolution, but have a lower level of prompt understanding, the model is harder to finetune. noobai stands out with the version that uses the v-pred training method instead of the traditional eps, with a larger amount of data than both pony and illustrious combined, it has an amazing level of diversity and prompt understanding, but the resolution is lower than illustrious, the v-pred version is quite difficult to train and use, but they also have an eps version
@DucHaiten Ok, thank you for all these details, it's very instructive and it makes me want to try it!
I would like to train loras for the Snapvu version.
- Should I train loras on noobai 1.0 eps or vpred?
- Any special training arguments for kohya?
- Dataset considerations?
I haven't trained lora or a finetune model on the base model with Snapvu, but I think it probably won't be a big problem, since it's still an eps model like the others, maybe there will be a difference maybe not, you can try with a small dataset first and tell me the results
So what is NoobAI exactly? Is it a variant of SDXL like Illustrious?
Noobai pony illustrious, are all models retrained from scratch with SDXL architecture, you can simply understand it as houses being demolished and rebuilt on the same piece of land. The finetune or dreambooth models are just expansions or interior and exterior decorations for the model that has been completed. Noobai pony illustrious, different in the huge data they train, how they train, training method...etc
the most obvious difference is that Pony aims to divide quality based on scores to highlight the training data more, with the least amount of training data. illustrious has more training data than pony, the generated images have the largest resolution, but have a lower level of prompt understanding, the model is harder to finetune. noobai stands out with the version that uses the v-pred training method instead of the traditional eps, with a larger amount of data than both pony and illustrious combined, it has an amazing level of diversity and prompt understanding, but the resolution is lower than illustrious, the v-pred version is quite difficult to train and use, but they also have an eps version
@DucHaiten thank you for the explanation. I appreciate it.
Any plans for EPS version of Cinematic? <3
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.