The model we have created is a combination of several general-purpose models, quite a significant number of them, so I don't believe listing a recipe or similar information is necessary (this means I don't really know the overall final type of license of this model, therefore, I propose that we utilize the original license as a starting point. Don't quote me on that, tho).
Without any specific style words or mentions of artists, this model generates in a semi-realistic, 3D-like anime-ish style.
Description
FAQ
Comments (11)
Looks nice. I'll try it out. It looks like it has some OrangeMix in it, but not sure what all else.
Btw, if you named it according to the Japanese word 雀蜂, that's an instance where the h becomes voiced and is converted to a b, making it Suzumebachi. That happens sometimes depending on the vowel just before it. It's basically the same thing as how in English we often pronounce t's as d's, like in "little."
Yeah, I know, but "suzumebachi" is also a character in Naruto or (and?) Bleach, so I make some arbitrary devoicing for uniqueness' sake.
@dobrosketchkun Ah, I understand now. Pay no attention to me then
I have no idea about Japanese, however pronouncing little as liddle is a bad example, as that's actually supposed to be a t sound in both American and British English. See https://dictionary.cambridge.org/us/pronunciation/english/little or https://en.wiktionary.org/wiki/little#Pronunciation
@anonderpling842 Yes, little is a bad example. A better one would be how in- changes to im- before words beginning with a plosive like b or p: ineffective but impossible.
However, the New Oxford American Dictionary gives the (only) correct pronunciation of little as | ˈlid(ə)l |. Also, according to the Wikipedia page on IPA pronunciation in English: "some dialects have a sound change known as intervocalic alveolar flapping, in which /t d/ are both pronounced as an alveolar flap [ɾ] between vowels or liquids and when not at the beginning of a stressed syllable." That alveolar flap is the consonant given in Wiktionary's second and third pronunciations and the one the speaker uses in the General American audio pronunciation.
Oxford Dictionary of English (for UK English) gives the pronunciation as | ˈlɪtl |, the same as the first pronunciation Wiktionary gives.
Can handle both semi-realistic and anime prompts, nice
This is an interesting and versatile model.
If possible can you also add the smaller 2gb pruned-fp16 version ?
This is the smallest weight I could achieve in the "checkpoint merger" tab in a1111. If you know another approach, I'm all ears.
I'm not an expert, just collecting models and learning from other users.
If this can be of any help:
- https://github.com/Akegarasu/sd-webui-model-converter
"convert to precisions: fp32, fp16, bf16 - pruning model: no-ema, ema-only - checkpoint ext convert: ckpt, safetensors"
- https://github.com/arenatemp/stable-diffusion-webui-model-toolkit
"Cleaning/pruning models - Converting to/from safetensors - Extracting/replacing model components"
And this is what another user said, in another model page in the comments section:
If you're using automatic1111 webui, you can do the following:
Go into checkpoint merger
Input this model as Model A (nothing else)
Name it (Something like "ModelMerge_Pruned_fp16")
Select "No interpolation option"
Select "safetensors" checkbox
tick the "save as float16" checkbox
Merge
What you'll get is a .safetensors file which is more secure than ckpt, 2GB~ in size and the generation is exactly the same as the .ckpt file.
@ritcher1 done
@dobrosketchkun Magic ! Thank you very much.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.