Hassaku aims to be a anime model with a bright and distinct anime style.
My Discord for everything related to anime models and art.
If you'd like to support my work, you can do so at: SubscribeStar, every bit helps and is truly appreciated!
____________________________________________________________
Supporters:
Thanks to my supporters tamashiicolle, pttcot, SETI and Kodokuna
____________________________________________________________
Using the model:
The model was trained primarily on images with minimal disruptive elements, such as floating text, logos, speech bubbles, or signatures. If any of these elements appear in an image, please include "signature" in your negative prompt.
Danbooru Metadata and franchise tags are not used in training, so please avoid using them in prompts. Metadata such as "highres" and franchise tags like "re:zero kara hajimeru isekai seikatsu" are not used in training.
A CFG scale between 3 and 7 is recommended. For best results, use CFG 6 with the Euler a sampler.
Refer to the example images, to understand, how the model should be used.
The model is designed to be simple and straightforward to prompt. Use the following tag order:
Number of characters
Character Tags
All remaining tags
Example: 1girl, rem \(re:zero\), standing, masterpiece, upper body
Here are some resolution options for SDXL:
1536 x 640
1344 x 768
1216 x 832
1152 x 896
1024 x 1024
896 x 1152
832 x 1216 (most recommended)
768 x 1344
640 x 1536
Lora use:
NoobAIs Loras working for the most part better in compare to Illustrious loras
Versions 2 and 3 is less compatible to most loras trained on Illustrious or Noob, because it got further trained away of both. Style A is still the most compatible for it.
______________________________________________________
Version and License info:
Below V1 merges ANIMAGINE XL 3.0
Version V1 use Illustrious-XL & WAI-NSFW-illustrious-SDXL with additional training
V2 is trained on its own and don't include any extra merge. Base was V1, it is trained to include newer or missing characters. It was also used for further training tests
V3 is a merge of Illustrious-XL and V2.2, to fix some issues and was also trained to include newer or missing characters
All Models using the Fair AI Public License 1.0-SD license.
Description
Update, mainly focused on fixing issues from v3.3
Better:
- Fewer artifacts and less face/eye melting, especially at a distance
- Cleaner, less blurry outputs
- Additional training on the same dataset af v3.3, some characters working now better
What i noticed in 3.3 + now 3.4:
- Hair color prompts are less reliable (eye colors gets used for the hair, as example)
- Skin can appear extremely bright/shiny again
- Multiple views occurs more often
I believe these issues started after training with Prodigy, it is maybe to agressive. I’ve since switched to ADOPT, and results improved a bit. I may release another update with further fixes and an updated dataset. Problem is that I’m still unsure when Z-Image or Klein will get a proper anime model. I also have several experimental datasets I want to test train, including mixes of natural language and tags (including NSFW, tags combine with natural language, that describe spatial awareness of characters and positions of texts, the exact weakness of tag based models).
FAQ
Comments (46)
Hassaku XL (Illustrious) 3.4 wont work with my loras
It works significantly worse than version 3.3.
Why did you delete version 3.3? Bring it back.
What works significantly worse? 3.3 is still available
@Ikena Thanks for bringing it back. Based on my personal experience and comparison with version 3.3, the results were significantly better with version 3.3 compared to 3.4.
@Gipno what promts did you use? My goal was to fix some of the ugly outputs of v3.3, so hearing that it performs significantly worse is concerning, More details would be helpful, please.
@Ikena I used the same hints as in version 3.3 and got worse results. I'll try to improve the hint.
@Gipno bro's trying to get feedback and maybe help and you're being vaguely subjective.
What makes it "worse", is it worse hands, adherence, consistency, accuracy?
The model seems more picky when lora is stack up but yes, v3.4 do fix some problem in 3.3. A way around some issues in this version is use SGM Uniform plus RKsampler (A111)/Adaptive ODE (Comfy) to address some noise left behind and be careful about lora stacking. This version is more like working with reactors, if you put time to experiment, the quality is really good.
Wasn't a fan of 3.3 but 3.4 seems great to me though.
I love this
Could you upload v3.4 on Tensor?
I'm not sure what happened to 3.0 before I could put at least 4 loras in there and I'd get perfect results, but now it really deteriorates even at one.
Use if possible noob loras. More training makes my model more and more its own thing. Like what illustrious to noob is. Loras will get less compatible. Older models like Style A are still available to stack loras.
@Ikena Thank you! This actually was extremely helpful :)
我喜欢这个
hey i gave 3.4 a test but as in 3.2 or 3.3 the eyes are sometimes completly to far away and this looks not good so i dont know which model u used to finetune ur base model pls check it again what the finetune model does because the eyes are not right in position >_ < <--- eyes looks sometimes like this there is too much space between the right position style is nice .
wierd,even normal scene it trys to generate things not for work
Really Good Model
What, not even a Hasshaku-sama image for a previw of this model? I remember how disappointed I was then the SD1.5 version didn't even know the character despite having her name.
newfags don't even know the character, lurk more
你好大佬，如果我想要生成纯2d动画风格需要下载哪个模型比较合适
试试平涂风格底模+anime screencap的lora+对应工作室提示词或lora？
1.3styleA
great model
good model
May be poisoned. Images turn off, malformed, often too much rainbow colours which weren't part of the prompt... it worked perfectly before.
ye idk why people are downvoting. 3.4 seems to explode when you use loras.
@Wizzi Its less compatible to most loras trained on illustrous or noob, because it got further training away of both. Style A is still the most compatible for it
@Ikena oh ok makes sense. thanks for the info. Might be a good idea to add this to the "about" section. a lot of people use models naked but some pick models based on lora compatibility.
@Wizzi what lora’s do you guys use? We could prob train artist styles on hassaku v3_base
@SupremeLeaderKimJongUn for me, character loras, pose loras, clothing loras, anime style loras. usually i use 3 out of 4 of these types of loras in every gen.
@Wizzi could you link some of those Loras? Character and artist style Loras can be trained more easily than others since the datasets aren't that hard to make
@Wizzi Same for me with character loras. Don't use the Loras for style that often
Can I request Tamamo Aria from Fate/Samurai Remnant in the next training? 🙏
I currently train a klein 4b prototype model, so sadly can't
finally a good model not focus only on humans. this model understand non living object interaction with human, like someone inside a locker or opening a door,
did you create tags your self with mixed of other non anime and non human images also?
Which version did you use to get these results?
@KaleidBlood 3.4
在使用这个model的时候如果使用Illustrious的Lora再进行图片放大的工作流，只要lora 的参数不是0/1，就会出现巨量的”lora key not loaded:“重复，并最终导致内存问题，我不清楚这是我工作流的设置问题还是其他原因，请大家在进行相关放大工作流的时候留意。（其他时候这个model很好用！）
If I want to make a style Lora for 3.4 should I train it on 3.4 or 3 Base?
If, If you ever find yourself with nothing better to do, and you remember my words, please create a new version of 1.3B. I've been using it ever since its release.
Amanatsu models are the new version
@Kalderas u saved my life
good
Details
Files
hassakuXLIllustrious_v34.safetensors
Mirrors
hassakuXLIllustrious_v34.safetensors
hassakuXLIllustrious_v34.safetensors
hassakuXLIllustrious_v34.safetensors
hassakuXLIllustrious_v34.safetensors
hassakuXLIllustrious_v34.safetensors
hassakuXLIllustrious_v34.safetensors
IL-hassakuXLIllustrious_v34.safetensors
hassakuXLIllustrious_v34.safetensors
hassakuv34.ILUSTRIOUS.safetensors
hassakuXLIllustrious_v34.safetensors
lyly-v2-02.safetensors
hassakuXLIllustrious_v34.safetensors
hassakuXLIllustrious_v34.safetensors
h4554kuXL_111_1U.safetensors
hassakuXLIllustrious_v34.safetensors
hassakuXLIllustrious_v34.safetensors
hassakuXLIllustrious_v34.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.