A 1/2 and 1/2 merge of my Western Toons Illustrious merge, which is mainly Hassaku XL + Jinkies Noob + PersonalMerge and Reynai's Oops! All Furries merge which is IndigoDream XL + IndigoFurryMix XL + Homo Simile XL + Tundras Mess XL 2.0 and a LoRA I trained on 600 EasyFluff 11.2 gens. The result is essentially EasyFluff 2.0, the penultimate furry toon Noob/Illustrious-based checkpoint.
All sample images are genned with no artist tags and no negatives or embeddings. This model will obey danbooru and e6 tags, but it has a very EasyFluff-esque house style.
Euler a, automatic, CFG 4.5 and 30 steps recommended.
Also recommended are the standard sdxl base resolutions:
1024x1024
1152x896
896x1152
1216x832
832x1216
1344x768
768x1344
640x1536
1536x640
Description
NAI version of Oops All Toons. This version has:
better prompt adherence
better eyes
less mis-figured hands
more saturated colors and sharper details
better e6 artist tag recognition
better compatibility with LoRAs trained on NAI EPS 1.1
Looser posing.
4.5 CFG, 30 steps, Euler Ancestral and SGM-uniform sampling reccomended.
Recipe:
Jinkies Noob
FAQ
Comments (10)
New version? Wow, what a great thing to see this morning! Please, keep cooking. You're an absolute godsend with these models and I love being able to see what newer things I can do with it.
Really getting some good things from this model. Thanks for sharing.
How does a LORA get mixed permanently into a checkpoint? I see you mentioned you added one based on 600 prior images you had
You can do this with either the Supermerger extension in A1111/Forge/Reforge or with simple merger in ComfyUI, just add a Load LoRA node before the save checkpoint node. Merged LoRAs don't behave the same when they're baked into a merge. Typically you only see an aesthetic difference in outputs, and trigger tags are lost. I like them for "driving" the style of a merge. For this reason I usually just bake in LoRAs that are unet trained only, since the text encoder training is mostly lost when adding to a merge.
Good
Why is the Illustrious version no longer a available?? Soo sad...
?
I've noticed that for some reason whenever I'm doing a face detailer with this model Recreate the entire image on the individual's face instead of just enhancing the features that are there.
It is a face detection problem. I don't know what webui you are using, I'm going to take the necessary steps in Forge; I think it is the same for A1111:
- You display the Adetailer tab.
- 'Detection'-> 'Detection Model Confidence Thresold': By default it is 0.3, increase it to 0.6 or 0.7. This causes Adetailer to detect only faces of a certain size.
- 'INPAINTING'-> 'Inpaint Mask Blur': By default it is 4, increase it to 8.
I hope this configuration works to you.
Quite possibly my favorite checkpoint to date. Bare minimum within my top 5 of all time... soooo many loras I used to create what this checkpoint does without asking! LOVE
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.