✦ Arthemy Comics Anima ✦
After a few attempts, I've managed to convert all of my tools for Anima and I'm so excited about this new base.
I've made this model by mostly "Removing" things from the Anima Base model, using a few of my LoRAs and modifying the model's internal values - like it would be done with a fine-tune training - by isolating concepts and boost them in order to highlight specific aesthetics or areas of the model.
Since I've worked on Qwen too, I highly suggest to download the tuned TextEncoder and use it with my model.
This means that this version is much more focused on 2D illustrations than the original model, and its scope is pretty limited to Comicbook / Cartoon outputs.
Said so, I hope you have fun!
___________________________________________________________________________________
1_____Size: Around 1024 x 1344
2_____Prompt: Start from the prompts present here and tune the keywords based on your needs:
Quick Prompt
score_7, score_8, dynamic pose and angle, extreme perspective deformation, solo, realistic western comics style, flat color 2d illustration, (insert your prompt here)
Quick Negative Prompt
no lineart, score_1, score_2, score_3, score_1, score_2, score_3, 3D, (insert your prompt here)
3_____CFG: 3.5 to 8.0
4_____Steps: 30 to 40
I hope you like my models and I can't wait to see what you all are going to create with it!
Description
FAQ
Comments (27)
The GOAT! I was sure this would come soon.
What's your thoughts about the Anima model, compared to Illustrious or others? I mean, which are your favourites?
I've played with a lot of models but it truly depends on the "scope" you have in mind.
Since I'm an illustrator and I really care about 2D illustrations, there aren't many competitors other than Anima for that spot right now - in all honesty illustrious is showing its limits and my training process was showing the first signs of diminishing return.
For illustrations I think that Anima right now is the best we could hope for - it's flexible, light, it can manage both Keyword based prompt and natural prompting.. it's really great!
I'm sure that, if the community starts to work on it like we did with illustrious, it can go really further than the SDXL base.
I mean, this model is the first iteration, made in a few days without experience on it and it's already showing great result.
The fact that you don't have a 75 token limit before it all goes down the toilet, alongside the fact that you have (keyword:5) (weights:2.5) for once in a blue moon, is quite fantastic.
That you can additionally do things like
"multiple panels, blahblah blah ..
right: something happens
left: something else happens
"
and it actually prints it, incredible
it also trains super easily, and very well
@Arthemy Thanks for the detailed insight! I tried it and it does feel flexible with prompting, understanding it most of the time. My only gripe for now would be the quality and consistency, compared to Illustrious that is. Compositions look pretty dynamic though.
Also, when you say you're an illustrator, you mean traditional art or AI? Just curious, being an artist myself who isn't scared of AI.
King of western style checkpoint is back !
It's still the first iteration with a model I'm not yet comfortable with - low expectations please! 😝✨
Of course , there is always a first step ;-) . This first version tend to "masculinize" many of woman's lora.
@Baekdoosixt unfortunately yes, as I said, not being familiar with the model, I might have pushed too far towards the masculine side, because I was used to fight back illustrious tendencies to generate female bodies.
For now my main objective was to try and westernize Anima a lot and get some feedbacks on areas that needs to be improved. ✨
@Arthemy that's a very promising first shot ! if i can be of any help please let me know
nice to see thisYou seem to be missing the "base model" category on this model, which is (unfortunately) rather important to find a n y t h i n g in the flood of prompt skill issue LoRAs for Anima.
Thank you for the tip! it was fix'd! :3
ITS TIME
I'm glad to see you've started working on Anima. It's definitely a model with the potential to improve Illustrious's work. That said, for the moment, Illustrious still produces generally better images (wider knowledge of Danbooru tags, precision, adaptability, style fidelity) than Anima, which is to be expected for such a new model that lacks years of training compared to ILXL models.
Back to your your checkpoint release, it has an install issue for Forge Neo users through Stability Matrix so i presume that this is the same for all Forge based tools:
CHv1.8.13: Run get_model_id_from_url CHv1.8.13: Getting model info for: 2700278 CHv1.8.13: Request model info from civitai: 2700278 CHv1.8.13: Requesting Civitai: https://civitai.com/api/v1/models/2700278 CHv1.8.13: Get subfolder for: G:\Stability Matrix\Data\Packages\forge-neo\models\Stable-diffusion CHv1.8.13: Searching local model by version id CHv1.8.13: folder: G:\Stability Matrix\Data\Packages\forge-neo\models\Stable-diffusion\sd\anima CHv1.8.13: model_ids: {'model': 2700278, 'version': 3032338} Traceback (most recent call last): File "G:\Stability Matrix\Data\Packages\forge-neo\venv\Lib\site-packages\gradio\queueing.py", line 536, in process_events response = await route_utils.call_process_api( ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ...<5 lines>... ) ^ File "G:\Stability Matrix\Data\Packages\forge-neo\venv\Lib\site-packages\gradio\route_utils.py", line 285, in call_process_api output = await app.get_blocks().process_api( ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ...<11 lines>... ) ^ File "G:\Stability Matrix\Data\Packages\forge-neo\venv\Lib\site-packages\gradio\blocks.py", line 1923, in process_api result = await self.call_function( ^^^^^^^^^^^^^^^^^^^^^^^^^ ...<8 lines>... ) ^ File "G:\Stability Matrix\Data\Packages\forge-neo\venv\Lib\site-packages\gradio\blocks.py", line 1520, in call_function prediction = await utils.async_iteration(iterator) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "G:\Stability Matrix\Data\Packages\forge-neo\venv\Lib\site-packages\gradio\utils.py", line 663, in async_iteration return await iterator.__anext__() ^^^^^^^^^^^^^^^^^^^^^^^^^^ File "G:\Stability Matrix\Data\Packages\forge-neo\venv\Lib\site-packages\gradio\utils.py", line 656, in __anext__ return await anyio.to_thread.run_sync( ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ run_sync_iterator_async, self.iterator, limiter=self.limiter ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ) ^ File "G:\Stability Matrix\Data\Packages\forge-neo\venv\Lib\site-packages\anyio\to_thread.py", line 63, in run_sync return await get_async_backend().run_sync_in_worker_thread( ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ func, args, abandon_on_cancel=abandon_on_cancel, limiter=limiter ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ) ^ File "G:\Stability Matrix\Data\Packages\forge-neo\venv\Lib\site-packages\anyio\_backends\_asyncio.py", line 2518, in run_sync_in_worker_thread return await future ^^^^^^^^^^^^ File "G:\Stability Matrix\Data\Packages\forge-neo\venv\Lib\site-packages\anyio\_backends\_asyncio.py", line 1002, in run result = context.run(func, *args) File "G:\Stability Matrix\Data\Packages\forge-neo\venv\Lib\site-packages\gradio\utils.py", line 639, in run_sync_iterator_async return next(iterator) File "G:\Stability Matrix\Data\Packages\forge-neo\venv\Lib\site-packages\gradio\utils.py", line 801, in gen_wrapper response = next(iterator) File "G:\Stability Matrix\Data\Packages\forge-neo\extensions\Stable-Diffusion-Webui-Civitai-Helper\ch_lib\model_action_civitai.py", line 847, in dl_model_by_input for result in download_files(filename, folder, ver_info, headers, filetypes, dl_all, duplicate): ~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "G:\Stability Matrix\Data\Packages\forge-neo\extensions\Stable-Diffusion-Webui-Civitai-Helper\ch_lib\model_action_civitai.py", line 644, in download_files dl_info = parse_file_info(ver_info, filename) File "G:\Stability Matrix\Data\Packages\forge-neo\extensions\Stable-Diffusion-Webui-Civitai-Helper\ch_lib\model_action_civitai.py", line 598, in parse_file_info filetype = file_info["type"] ~~~~~~~~~^^^^^^^^ KeyError: 'type'
I've investigated the issue and it doesn't seem connected to my model but it's about how ForgeNEO is currently loading the Anima models. If you ask an LLM about it, it will explain you how to fix it... maybe.
@Arthemy Interesting, i ll take a look at it. That being said your model is the only one causing this issue among my Anima ones. So, whatever any LLM might say, i m rather skeptical about the cause not being connected to this model.
@antarek I think I might have spotted the problem:
The original model was saved with a custom node of mine which has the ability to "bake" model patches inside a model, in that way I can "tune" different slices of the model and modify them as much as I want - the problem might have been caused by an header mismatch that made Forge-Neo cause an error.
I re-uploaded the model, which I just re-saved with a standard node - it "should" work as intended now! :D
@Arthemy Manually downloading the model and using civitai helper to download this model infos did the trick. So if anybody else happens to face this issue then this is how to solve it.
After playing around a bit with your new Anima model, first, i must congratulate you on what it already allows us to achieve. You never disapoint, hats off. I noticed an issue tho using it on Forge Neo. It tends to try to force patterns or jpeg artifacts on pictures when using img2img (at a denoising strength of 5 most of the time). It seems to be somewhat seed related and setting "jpeg artifacts" tag score to 2 and adding "patterned" with a tag score of 2 seems to help in your negative prompt.
jpeg artifact issue: https://i.ibb.co/Pv8QtPK6/jpeg-artifacts-issue.png
pattern issue: https://i.ibb.co/S7HNZw3V/pattern-issue.png
@antarek I took a few months to reach the quality of my "Arthemy Western art" I will take time to improve this model and get comfortable with it; but these feedback are exactly what can help me achieve a better result. It's always a group effort to find blind spots and mistakes - thank you! ❤️ I'll tackle this issue with the next version.
@Arthemy After further testing, and to confirm my previous comment, this behavior is occuring when your model wants to represent something specific but can't do it because of your negative prompt. For exemple if i want to generate a picture of a character fighting or running if your model wants to place sweat related stuff on the skin of your character but can't because of having "sweat" and "sweatdrop" into your negative prompt it ll force patterns into spots where it would have placed sweat-related stuff. And, of course, this is seed related. Some ll produce that model behavior and some won't.
This model is the best, but somehow all the characters seem angry even when prompting other expressions. Anyone else has this issue?
Not on my side. What tag score range do you use? I m asking because tag scores on Anima aren't the same than on XL models. It handles a way larger range (tested on tag scores set to 10 with no burning). So try using whatever facial expression you want to use associated to a score of 2, 3, 4, 5, etc...
@antarek Hey, thanks for replying, i tryed all scores after you're comment the characters are still angry, I think this has to do with the cocktail of style loras i use because when i remove them i can get diffrent expressions, still this doesn't happen to me with any other chekpoints.
@panico0o Then you are probably on the right track. The less LoRA the better anyway. For styles you have this amazing web tool that allows you to see most styles currently supported by Anima base1 so you can paste any of those styles trigger into your prompt. This is for anime styles but neither Illustrious nor Anima are western comics focused sadly. Thus why having model creators as talented as Arthemy is a true blessing for people that enjoy the western art style over the anime one.
@panico0o feel free to share the images you're talking about, with all the settings, I'll do my best to analyse them and check if there is something wrong with my model. Anyone is welcome posting errors and bad looking images - it's the only way I can get a better idea of what works and what's not working. ✨