@asamiarts tracing AI
Best viewed on PC, the formatting is broken on mobile and image descriptions might not display on the intended places.
There is a big issue at the moment with artists tracing over AI while charging hundreds of dollars for commissions in platforms like VGen
This doc is regarding @Asamiarts (formerly @Drawingasami) and it's meant to inform commissioners about ongoing tracer issues. I don’t like making docs like this, I don’t like accusing people of AI and I hate scrutinizing other’s art or questioning their skill. But at the moment, Asamiarts is taking $120 commissions through VGen while blocking anyone who questions if they use AI. I understand it’s very frustrating to have people constantly doubting your artistic skill, especially when it leads to unnecessary insults or harassment. In this doc I will show you why I believe the odds of Asamiarts tracing over AI are extremely high. Generally, there is always a pattern when people do this. The behavior these tracers exhibit always ends up being very similar.
Some AI account red flags
- The Twitter/X art account is very new (Most Joined after 2021) and there are no other socials associated with that username or similar This is not the case with Asamiarts as @Asamiarts Joined August 2024, but they had a previous account with the username @DrawingAsami (Joined 2022) that was suspended.
- Art is constantly deleted when someone points out there are AI elements, or comments mentioning AI are ignored and hidden.
- There are multiple sudden changes in art style in a short amount of time. Please note it is completely normal for artists to study, experiment and to draw in different styles, but in cases like this it’s kinda drastic and there is no prior practicing leading to the change. One day there is a drastic shift and they have a different style from an existing artist.
- In Asamiarts case, it combines both of the previous points: a lot of their previous art has been deleted because they suddenly switched art styles from popular japanese artists who have AI models trained with their art without their consent. I will elaborate more on this later. Their current twitter media tab is all art that resembles famous jp artist Ixy（いくしー）, they have deleted all the art that resembled 荻 pote (ogipote).
- Sudden jump in art skill. For example, an amateur digital artist suddenly mastering anatomy and color theory within a few months. You can’t observe this simply by looking at Asamichan’s media tab because they have deleted most of their old art from twitter and pixiv. I will be showing you their previous art before their current Ixy style. I have to mention in a lot of these cases the person tracing will still struggle with lineart because you can trace the anatomy and copy the colors and shading of an AI picture, but the AI will not magically improve your lack of experience doing linework. Unfortunately this point doesn’t apply when the image is being traced by a veteran artist, because professionals will have good linework control already.
- Refusing to provide proof, deflecting, ignoring people. Once it becomes too overwhelming, tracers provide a small amount of evidence that ends up raising more questions, and most of the time they also delete that evidence. Proving that you aren’t tracing is really simple if you are NOT tracing AI. This person was invited to draw together in a browser-based canvas website but their response was that they don’t have to prove themselves to anyone or convince anyone. The website in question being magma.com, a site similar to iScribble or Drawpile but online. This is a simple, free way to show everyone you are not tracing.
TIMELAPSE INCONSISTENCIES
The timelapses below are from Clip Studio Paint. CSP allows you to export timelapses in 15, 30, 60 sec lengths, and a “full” length. I converted the original .mp4 files that were posted on twitter to .gif format so you can view it here. The timelapses that Asamiarts posted were 30 seconds long, and the gif version is also 30 seconds long.
The lineart magically being drawn with no sketch in the background
This is how CSP timelapses work: You have to manually press the “record timelapse” button, then the program will begin recording all the strokes and undos you make in every layer. Sometimes artists forget to press the record button so the timelapse won't show them working on the sketch, but it will still show the sketch at a low opacity when they are lining. This means Asami is lining without a sketch to guide her, without corrections, perfectly placing every line with no hesitation.
Note how she doesn’t even bother drawing Calli’s (right timelapse) whole face first then erasing the chin area that will be covered.
You should know that Clip Studio allows you to make a private linework layer that will not be shown in the timelapse: Clip Studio Paint Has Private Layers?. I have included the source links and original upload timestamps for Asami’s timelapses, but as of October 9, both timelapses were deleted from her account. Below I will show some timelapse gifs from AI tracers, and a twitter user who traced an AI image to prove that it’s possible to fake a timelapse in CSP. These timelapses are very similar to Asami’s where there are no adjustments and the linework is drawn perfectly.
🡅 Mashiro_www tracing with Procreate 🡅 Misouuy tracing with CSP, read more here
🡄 The timelapse to the left was posted by a
twitter user to show you can trace over AI
with CSP and it won’t show in the recording
(12/2 Update) To see a comparison with timelapses by other artists, check this separate document.
And once more, please note that there is no “standard” timelapse process, as all artists work differently. What stands out most of the time between real and fake ones is usually how perfectly the pose draft appears at the start, there’s no sketch phase shown, no adjustments, etc but this is not the norm. It also depends on the artist’s experience.
On October 7 , when I QRT’d Calli’s timelapse commenting about it being traced from AI, Asamiarts immediately blocked me. On October 8 she posted a new timelapse, then deleted the two old timelapses on October 9 . Both deleted timelapses had been in her media tab since April 2025 before that.
What did Asamiarts have to say about this? She was contacted via VGen regarding proof of her art not being traced over AI during October 8.
When questioned about how she managed to do the lineart without a sketch this was her reply.
This is the direct reply to the purple message above.
I thought for a second that maybe english wasn’t her native language and that’s why she typed it that way, but the other messages are not written in the same vague style. It’s basically a non-answer.
As she mentioned in the first message, she posted a new timelapse during October 8. 🡆
Before I elaborate on this third timelapse let me finish this exchange that happened on VGen.
https://x.com/Asamiarts/status/1976045767598157959
🡄 Asami is claiming here that she erases and redoes her drawings many times “but Clip Studio only records the video from where you last left off to speed up the drawing process.” This is NOT TRUE. After you press the Record timelapse button, the only thing you can do is either export it or press the Record timelapse button again to stop recording, and CSP will show you this warning if you try to do that.
🡅 Not shown in the exchange above is a previous message mentioning
she’s been drawing since 2017.
Now let me elaborate on the third timelapse she has posted on her twitter. This is no longer a Clip Studio timelapse, this is a Procreate timelapse. Procreate is a popular iPad drawing app, which includes a timelapse export feature just like CSP. However, Procreate is known for having a feature where you can easily hide entire layers from showing in your timelapse. With CSP, you can only hide linearts or binary images. Sadly because of that, I do not consider a procreate timelapse to be solid evidence in a case like this. If you want to see what this looks like in action, please view the following short: How to hide your reference layer in Procreate.
I’m sorry, but I cannot take this as proof that Asami isn’t tracing because it was recorded in Procreate. She also provided a photo of the layers, but you can delete the traced image layer after you are done tracing. And again, a day after posting this timelapse on twitter, she deleted the two previous ones. I don’t want to be nitpicky, but even in this new procreate timelapse, she never goes back to erase, adjust or correct anything while sketching, lining or shading.
October 2025 timelapse (Posted October 8) April 2025 timelapses (Deleted October 9)
There is something to note about how the improvement rate becomes static from April to October, but I will elaborate later.
I believe showing layers is only solid proof for an artwork that seems to be generated with AI, but it’s different when it’s someone tracing over AI. Clearly if the image is gen AI then the artist won’t have layers to show (or they will claim that they drew everything in one layer). But I believe Asami is tracing, not posting gen AI (However, there are some previous artworks I believe were generated with AI and not traced)
So what would someone do in a situation like this, where the timelapses are weird and you are still being accused of tracing AI? How do you prove your art is legit?
On October 10, it was forwarded that I'd be willing to draw with Asami on magma.com in real time. Magma (previously known as Aggie) is basically a browser based drawing app, it functions exactly like any other art program. If she isn’t tracing, then she would have no problem drawing with another artist at the same time (however, it’s understandable to be uncomfortable if it’s a random person). Her reply to the purple message to the right is below 🡇
You might not owe me anything, but you do owe your current clients the right to know what they are truly paying for. No one has to commission you, but the people who ARE commissioning you are doing it because they expect to receive an illustration that was drawn by you with your own skills, and not something you generated and traced over. Just like last time, this isn’t even specifically about AI art, tracing and plagiarizing has always been a thing in the art community. AI has just made this issue way more common because it’s easy to just generate an image and delete it once you’re done tracing it, then the evidence is gone forever. But in the old fashioned way, if you are tracing an existing artwork from another artist, the original source can be found online eventually.
One day before I saw Asami’s reply, someone I know reached out to another artist because they seem to be tracing over AI as well. It’s best to clear up misunderstandings about AI in private, but this was the reply that the person I know received, which reads very similarly to Asami’s. It also mentioned drawing in Magma.
🡅Message from person A (Anonymous)
🡅 Reply from Person B (AI tracer)
For the record, I also believe that person is most likely tracing but unlike Asami they are not taking commissions, and this doc isn’t about them. However, the similarities are there and I found the timing to be astonishing. Most of the time AI tracers will immediately deflect or become hostile when people start pointing out the issues and inconsistencies with their “art”. The screenshot on the right is from @foame_/@misouuy, replying to someone on twitter DMs after people questioned if they were using AI.
REFUSING TO PROVIDE EVIDENCE WHEN QUESTIONED
Asami’s current media tab is missing a lot of her art since the posts have been deleted, most of the time because people pointed out AI inconsistencies. Her previous account @drawingasami was suspended (I don’t know the reason for the suspension) but even back then (2024), people were suspicious of her using AI.
Source https://danbooru.donmai.us/posts/7994721 Uploaded to Danbooru on August 2024 | Original Pixiv upload https://www.pixiv.net/en/artworks/121451763 (Deleted)
🡄 The comments on the left are talking about this artwork Asami had posted in her previous account (May 2024). Notice how different the art style is compared to her current style, I will talk more about that later. I was not able to recover any of her replies as the account was suspended.
(Danbooru reupload)
The comment below was a reply to the Mumei art wondering if it was traced
AI ERRORS IN THE ART
I try to not be nitpicky about errors in art, because human errors are a thing too and I have seen people accuse others of AI because of mistakes in their art that can be attributed to human error. It only becomes so evident when these mistakes happen on almost every single image, to the point all of them have at least one issue. In fact there’s so many things to point out on every artwork that it would take too many pages, so I will point out the most blatant ones only.
Gun muzzle placement
Here’s Asami’s explanation for this gun she drew.
https://x.com/Asamiarts/status/1915693130625536326
I’m not an expert at drawing guns, so when I draw them I use tons of references or 3D models. But it’s just humanely not possible to make a mistake like this if you even bothered to look up a single handgun on google images. I don’t want to ramble about this but I simply can’t imagine looking up references for a gun and then placing the muzzle on the rear of the gun (ignoring the lack of proper finger placement). There are thousands of images and 3D models of handguns online, if you google handgun refs you will draw an existing gun like a M1911 or a Glock.
You can skip this page, I just want to emphasize how ridiculous this error is. This is what I mean when I say AI users do not care about details or are unaware of their mistakes. You really saw the muzzle in the back and didn’t think for a second that this guy is gonna shoot himself. (Assuming she doesn't know gun parts or how guns work, at least you’d think that the big hole in a weapon is where the bullets come out of)
Giving the benefit of the doubt, maybe Asami has never seen a gun in her life or played a videogame with guns, she has no idea what's back there. And if you google “handgun” for references the first results only show them lying flat. But you would assume she would have also googled something like “man holding a handgun” to properly draw the hand in the image, and there are plenty of images with the right perspective and no back muzzles.
Note that Asami’s post explaining why the gun looks the way it does wasn’t posted right after she uploaded the artwork on April 24, it was posted after the comments on the left where they pointed out the same issue.
I’m not a gun expert, but if such a handgun with a hole in the back exists and it was used as a reference let me know. I’m not even being sarcastic, I know of a handgun that has a round indent in the back but I will refrain from sharing the name here in case Asami reads all of this and tries to pretend THAT was the gun she referenced from. But it’s a bit unbelievable she didn't use a more widely available reference when the most common handguns look like this (From left to right: SIG Sauer, Glock, M1911)
Hair and clothing errors
Pixiv source (DELETED) | Danbooru source (Uploaded Nov 2024)
It’s always Raora with these tracers man
The top part of the sweater doesn’t connect to anything, and there is a random string connected to the underwear (but you could say it’s just a fashion design element).
In this La+ image, there are hair pieces that are not connected to the main hair.
Pixiv source (DELETED) | Danbooru source (Uploaded Nov 2024)
Twitter source (DELETED) | Danbooru Source (Uploaded Dec 2024) [NSFW]
There are so many things that are wrong with this pic that it leads me to believe it was generated instead of traced. The hair is so disconnected and just lacks coherence everywhere. However, it is too jpeg artifacted for me to be able to tell if it was generated or not.
You will notice a lot of these issues will be hair errors, when the person who is tracing has no idea how hair strands work and is tracing over random AI hair chunks this is what happens (looping or disjointed hair strands)
Twitter source (DELETED) | Danbooru source (Uploaded Dec 2024) Pixiv source (DELETED) | Danbooru source (Uploaded March 2025)
From left to right
(ALL DELETED from twitter and pixiv)
https://danbooru.donmai.us/posts/9044187 (March 2025)
https://danbooru.donmai.us/posts/9088713 (April 2025)
https://danbooru.donmai.us/posts/9116902 (April 2025)
Raora double tail, Asuna errors
Raora either has two tails here, or the tail loops around and becomes thicker, or it bends to break around.
Danboooru source (Uploaded October 2024)
For Asuna, I suspect the image might be generated, not traced. Shadows do not match the shadow source or have different hues, hair chunks are not coherent, her ribbon has 3 bows coming out of it instead of 2.
Further AI errors can be spotted in the full resolution image
Twitter source (DELETED) | Danbooru source (June 2025)
Strap going under the shoulder instead of on top Ina hair chunks (you could argue that her hair is wet)
Twitter source (DELETED) | Danbooru source (July 2025) Twitter source (08/12 STILL UP IN THEIR TWITTER!!! FINALLY!) | Danbooru source (June 2025)
AI USERS IGNORE CRITICAL CHARACTER DESIGN ELEMENTS
Twitter source (DELETED) | Danbooru source (Uploaded May 2025)
Raora art, major design inconsistencies
What I have noticed when people are tracing is that most fanart they post has the character wearing simple casual clothes like gym clothes, plain dresses, swimsuits, etc. But when they post fanart where the character is wearing their official base outfits, design elements are wrong or do not match the official references. I find it very hard to believe these are human errors (for example: only using a low quality reference, or having to interpret details yourself because the character you are drawing doesn’t have a reference sheet available). Raora has a very detailed reference sheet available online, and plenty of fanart with the correct accessory shapes. There are also AI elements across the image:
______Raora reference sheet source
Kanata art, design inconsistencies
🡄 One could argue about the artistic liberties an artist can take while drawing an official outfit (for example and if you are familiar, Touhou Project characters). Maybe the switch from pantyhose to thighs was because the artist specifically wanted to show the underwear, however, there are too many elements that are also missing.
Sakuna art, design inconsistencies
🡇 In Sakuna’s case, the side bangs are a completely different shape, which is something AI tracers tend to ignore sometimes. I was not able to find a reference of her live2D model (only an official render) but you can see her model in her youtube channel.
Pixiv source (DELETED) Danbooru source (Uploaded Oct 2024)
‘
Pixiv source (DELETED) | Danbooru source (Uploaded December 2024) Sakuna Ch. Youtube thumbnail (Debut stream) Official art
Note: Artists might still interpret hairstyles in their own unique way to suit their style and it’s not always indicative of AI tracing.
The next example about a character not matching their official design will lead into my next point of AI that is trained with popular artists' works, emulating their style very accurately. In order to present that, I will quickly explain some info regarding AI models.
Have you noticed how during 2022 AI struggled to figure out character outfits and blended them together, but recently they are able to match character designs more accurately? This is thanks to LoRAs.
Basically, the LoRA is a smaller model that is trained with very specific data that helps fine-tune details in other larger models. This isn’t just done for character costumes only but also things like hands, poses and most importantly (in this case) specific art styles.
Screenshot source: What is LoRA (Low-Rank Adaption)? | IBM
Saba pre-debut artwork design inconsistencies
On June 28, 2025 the Vtuber Sameko Saba debuted. Lets just say Saba’s debut was highly anticipated by a lot of people, and tons of artists were eager to see her design and draw fanart for her. A few days before her debut, Saba started posting some teasers that showed some parts of her design, so fans were uploading their own design predictions and started working on fanart before the debut date. The earliest AI LoRAs I can find of Saba were uploaded June 24, 2025, trained with the artworks that were guessing what she would look like. Because of this, the design that was generated by AI with these LoRAs was inaccurate.
The June 24, 2025 upload times for
the LoRA models shown on the right
are 5:50:43 PM (EST) and 6:40:44 PM (EST).
On June 25, 2025, Asamiarts posted a Saba fanart. Keep in mind, at this point there were only teasers for Saba’s design. Everyone was guessing and coming up with their own predictions, so it would not be possible for any fanart to fully match the official design that would be revealed on June 28.
Twitter source (DELETED) | Danbooru source (Uploaded June 25 2025)
The reason I suspect this artwork is traced from AI is not necessarily because the design does not match Saba, as I have pointed out her full design was not revealed by the time Asamiarts posted this. Rather, I find it hard to believe this was their own design prediction by June 25, when the following teaser had been posted by Saba during June 15:
Video teaser uploaded on June 15, 2025 where you can see her tail shape and colors clearly. Note the blue dress hemline.
On June 22, Saba uploaded a final teaser that showed the back of her head and part of her side profile, along with a silhouette of her body. In this silhouette her tail shape is very clear too. You can also see she has no blue highlights on her side bangs. There is a very small probability that they missed these two videos, but I find it hard to believe because the new fanarts being uploaded had the correct tail shape drawn too. More importantly and going back to the LoRAs, the tail shape is clearly taken from another vtuber and one of the LoRA uploaders admits as much:
Shark vtuber tail in question
On June 25, Saba finally debuted and made her reference sheet available. After new fanarts started being drawn and uploaded, AI users immediately began scraping them to make LoRAs with her proper official design. The earliest models that I can find (trained with her design that was revealed on June 25) were uploaded on June 28. On June 30, Asami uploaded the following artwork:
Twitter source / Danbooru repost in case it gets deleted
Going back to AI users ignoring design elements, and despite the new updated LoRA for the outfit, once more there are a few things missing or incorrect in the artwork above, such as the tail position and length. The dress does not have a hole in the back for the tail, it's meant to be placed underneath it.
Saba official reference sheet Back view Side view
And back to LoRAs again, do know that multiple LoRAs can be used together. That means you can generate an image with the Saba outfit LoRA + Ixy style LoRA together. Who is Ixy? An artist whose illustrations have been used without permission to train multiple AI LoRAs and the style Asami has begun using recently. However, Asami has used multiple styles in the past, all specifically from Japanese artists (who also just happen to have LoRAs trained without permission). This will be the focus of my next point.
ART STYLE CHANGES ARE SUDDEN, OLD ART DELETED
Asamiarts has changed styles many times and wiped her galleries (both pixiv and twitter) on multiple occasions to hide this. We took a screenshot of her pixiv page on October 7, and while I was typing this during October 11 I went to check her pixiv and most of the art was gone, again.
October 6 October 11
On a personal note, I don’t find artists deleting art that suspicious unless there are other issues, as someone who used to delete art quite often (I would not be satisfied with the final image after posting it, or deleted old art so only recent artworks would be displayed in my media tab. Now thanks to AI, artists are forced to keep old art in their media tab as proof of improvement over time in case they get accused of AI). The concern begins when the art or “evidence” such as timelapses are deleted only shortly after comments start pointing out AI elements.
‘
(Replies to Asami’s uploads from twitter)
Disclaimer: It is completely okay for artists to experiment with different styles and that itself is not indicative of AI usage. But unless you are a professional, it’s not so easy to switch or emulate styles so quickly (less than a month) with no inbetween practice at all. And in this case, they all happen to be styles from well known japanese artists.
Ixy art style
The current style they are using is based off Japanese artist Ixy
Their art has been used without permission to train multiple LoRAs. But once more, you can combine multiple LoRAs together so it could be possible they are using the Ixy style LoRA with another artist style.
Asamiarts (July 2024) Ixy AI LoRA (April 2024) Ixy original (Nov 2020)
In order to respect Ixy I will not be sharing or linking any other of their artworks going forward. But I was not the first person who noticed Ixy’s style resemblance in Asami’s artworks during this time period (as mentioned before, she has changed art styles in the past).
Something very important to consider is that the style change needs to happen after an AI model/LoRA is uploaded online. If the LoRA was not available online by the time of the sudden style change I would not consider it a possibility for the art to be traced from AI, but I am also aware anyone can train their own models offline and not share them with anyone.
In Asami’s case, and this is just an estimate based on danbooru upload dates, the Ixy style change happened around June 2024. For example, the model to the right was published in August 2024, so I wouldn’t consider that it was used at that point in time (perhaps after that date). However, there are Ixy LoRAs dating all the way back to 2023 online, with 2024-2025 ones producing more accurate generations.
The following screenshot is a LoRA model reupload from an AI site, but the original upload is no longer available. So there is also the possibility that it was uploaded online much earlier than April 28 2024.
Some of the artworks uploaded more recently resemble pose generations using Ixy’s LoRAs to an extent. But since AI learns from real art, I try to not focus on pose similarities unless it’s a very close match.
ASAMIARTS AI GEN
Twitter source (deleted) | Danbooru source (July 20 2025) AI upload (June 8 2025)
‘
Twitter source (Sept 18 2025) | Danbooru mirror Twitter source (March 30 2025)
Obuchi314 uses AI with Ixy and BLADE’s art styles
Ogipote art style
The art style they used before was based off Japanese artist Ogipote
This is the style they used from 2023 to mid 2024. I will point out that Asami is also following both Ixy and Ogipote on Bluesky and Pixiv, as well as two other artists with similar anime-nuri styles. Ogipote has also had to deal with their art being used without permission to train AI models since 2023.
🡅Asamiarts (Jan 2024) 🡅Ogipote
Before May-June 2024, Asami was using Ogipote’s style before the switch to Ixy. Unless there are missing images in the multiple boorus I searched, this switch was sudden and drastic. I will stop listing twitter sources and only use boorus as all of the original posts for these are deleted.
From left to right: May 2024, May 2024, June 2024 , August 2024
My reason to believe these might also be traced from AI comes from the fact that there are also multiple AI elements in the backgrounds. In most instances the background is blurred to hide these artifacts.
From left to right (2023) 1, 2, 3, 4, 5
Sourced from Asami’s tumblr
I will not be showing every single instance of AI bgs since there are dozens of images. You can argue that only the backgrounds are AI generated, and the characters are drawn from scratch on top and not traced from AI. But the switch from Ogipote to Ixy style was just as sudden as the jump in skill before Ogipote’s style, which is what leads me to believe the characters started being traced one point too.
Once more, let’s look at LoRA upload dates. If there is no AI model that was uploaded online before the art was posted, then the chances that it’s traced from AI becomes lower. The earliest Asami artworks in danbooru with Ogipote’s style date back to at least June 2023 (April 2023 commission in Ogipote style was found later), but as previously mentioned, Ogipote tweeted about AI infringement all the way back in April 2023. The earliest model I could find dates back to March 2023 and can generate the background at the same time.
I believe Asami started tracing AI around this time period. During my research, I found out that one of my followers had commissioned Asami multiple times back in 2023. I reached out to them and explained the situation, and they provided me with the commissioned images and gave me permission to use them in this doc.
Out of the 3 commissioned images they shared with me, the one with Ogipote’s style was dated back to April 12 2023, which is earlier than the June 2023 danbooru upload date I mentioned above. However, my point still stands as the AI model for Ogipote’s style was published on March 22 2023.
🡅 Ogipote’s art being used to train AI with their style has been an issue since 2023
Color palette, art style, anatomy jump
These improvement rates in a single month from beginner artists only became so common after 2022 with the first wave of AI image generation. For non artist’s references of what general improvement looks like, see the Artist Evolution posts on Pixiv (one year art improvement meme) or Ten Year Progress summaries.
Eku Uekura art style
There was a small period of time where Uekura Eku’s style was used.
Around May 2024 (approximate according to booru uploads), Asami briefly switched styles from Ogipote to Eku Uekura’s art style before switching again to the current Ixy style.
Something I noticed is that while Asami is following both Ixy and Ogipote in Bluesky, Uekura is blocked. Asami’s bsky was created Feb 7 2024 and they were blocked on Feb 28 2024. Source: Clearsky
From left to right: June 2024, May 2024
This is strange as it’s not part of a mass block (Bsky lets you mass block accounts from lists made by other people)
🡅 Asamiarts (May 2024) 🡅Uekura Eku
Once more, let’s look at LoRA uploads online. At first, I could only find one uploaded on February 29, which is one day after Asami blocked Eku on bsky. In the end this similar timing was probably a coincidence, and I thought for a second that maybe the block was an accident. However, I find it unlikely as you have to press a menu to find the block button and you get a warning when you are about to block someone.
That said, after a bit more digging I found a model that was uploaded on October 24, 2023.
There will always be the possibility that these are simply heavily referenced, and not generated with AI and then traced. But somehow these exhibit similar AI errors that are still persistent in future artworks.
🡅 Source (May 2024) 🡅 Source (June 2024)
Visualization of style change
I haven’t gotten to Asami’s art prior to 2023 yet, but I want to share some comparisons of the style change with the same characters across less than a year.
March 2024 August 2024 January 2025
July 2024 September 2024 (discord embed source) October 2024 December 2024
Incoherent skill improvement
As observed, there is a constant and very distinct change in styles but besides that improvement completely stalls.
So far I’ve tried to not nitpick at errors that could be human in nature, but there are some issues that can’t be dismissed.
There are pieces where the perspective seems flawlessly executed but then there
are artworks with dynamic poses that have anatomy mistakes and proportions are always different between images.
May 2025 Source: Twitter (deleted), Discord embed shows upload date
If an artist has somehow mastered multiple styles in less than a year, seems to have knowledge of anatomy (to a fluctuating degree) and dynamic poses, with over 8 years of experience,
you would expect them to know how patterns wrap or bend around clothing and how to execute it in their art program of choice instead of multiplying a font layer and using the free transform tool with no account for the curvature of the body.
🡅Twitter source since this is still up (May 1 2025)🡅
(Nevermind! I typed that and confirmed it was up on Nov 28, but as of Nov 29
it has been deleted. Have a danbooru source dated to May 2025 instead)
ART CHRONOLOGICALLY
2021
MARCH 2021 ➞ SEPTEMBER 2021 ➞ ??? 2021
Sources: 1. DeviantArt web archive capture. 2. DeviantArt web archive capture. 3. Gelbooru, reuploaded by a commissioner in 2024. I DM’d them and asked when it was drawn and explained the situation. The screenshot on top of the image was their reply.
2022
MAY 2022 ➞ MAY 2022 ➞ JUNE 2022
Sources: 1. Etsy reviews. 2. DeviantArt web archive capture. 3. Commissioner reupload on Toyhouse, credited to Asami’s DeviantArt.
AUGUST 2022 ➞ SEPTEMBER 2022 ➞ NOVEMBER 2022
(Side note: Stable Diffusion Released on August 2022)
Sources: 1. Etsy reviews. 2. Commissioner reupload on twitter 3. Commissioner reupload on twitter
2023
JANUARY 2023 (AI backgrounds start showing up) ➞ JANUARY 2023 ➞ FEBRUARY 2023
Sources: 1. Commissioner reupload on FurAffinity credited to Asami’s Etsy, also seen in the Etsy reviews. 2. Etsy reviews 3. Commissioner reupload on twitter
MARCH 2023 ➞ APRIL 2023 (Ogipote style starts showing up) ➞ MAY 2023
Sources: 1. Etsy reviews 2. Commissioner’s
original file 3. Commissioner reupload on twitter
The first artwork’s review on Etsy mentions that the character does not look similar to the reference that was given for the commission. However, I don’t know what the original character looks like, so I’m unable to make a comparison.
2024
JANUARY 2024 ➞ FEBRUARY 2024 ➞ NOVEMBER 2024
Sources: 1. Danbooru 2. Etsy reviews 3. Danbooru
2025
MARCH 2025 ➞ APRIL 2025 ➞ JULY 2025
Sources: 1. Danbooru 2. Danbooru 3. Danbooru (Not pasting timestamps but they are visible in danbooru)
DELETING ART AFTER BEING CONTACTED
I was going to source the first image to her Pixiv, as I remembered it was still available there and you can see it in the Pixiv screenshot back here. It’s now gone, I believe it was deleted in the past few days. There is only one artwork remaining in her Pixiv now.
I checked twitter instead, and I found out that the remaining posts from early 2025 are all deleted now. Last night/early morning (November 28) I had been scrolling through her twitter for sources, so I believe this mass deletion happened overnight as I typed this (3am, November 29). All of the remaining fanart has been deleted, and now the media tab solely has artworks of her OC going back to September 2025.
I’m sorry I don’t have screenshots of her old media tab to prove this so you will have to take my word on this one. The only thing I can provide is the version history for this doc which proves I did check her twitter yesterday and the link was still working (I was surprised as all of the previous pics I had in the doc only had booru sources, which is why I updated the text mentioning it was still up.)
Do you know what happened before midnight Nov 28? I emailed Asami. So I am going to assume that this specific deletion spree was not a coincidence. In my email, I apologized for the harsh wording of my QRT back in October of her Calli timelapse, where I implied it was someone tracing over AI. Even if it looks most likely traced, I shouldn’t have replied the way I did. Of course, I was quickly blocked after she saw my QRT so there was no other way to contact her. After apologizing, I briefly explained the increasing amount of artists tracing AI, and expressed wanting to solve these matters privately. See, after taking a 2 week break from this doc I started to wonder if maybe I had just hallucinated everything, and maybe all the theories about the LoRAs were far-fetched as I have no way to prove she did use those models. Maybe the improvement is suspiciously fast yet disconnected but who knows really because it’s something that can’t be standardized.
That’s why I decided to email Asami to clear things up. Quite honestly I do not expect a reply. I asked for PSD files of any of these four artworks seen on the right, but I know an artist might be uncomfortable sharing PSDs with strangers so I mentioned screenshots of the layers were okay but told them that I know those can be edited. I didn’t pick these artworks randomly though. I have a suspicion these are generated, not traced.Yet I can’t confidently say they ARE generated because I have no evident proof and I could be wrong. They might simply look like AI because of the blur filter.
Why do I suspect these might be generated? I had mentioned the Asuna issues back here but I didn’t point out everything that seemed odd back then. Source for the images on the right: 1, 2, 3, 4
There are some errors where I think the possibility of them being human mistakes are extremely low, but in some cases it’s very ambiguous as there are genuine mistakes in art sometimes because the artist was distracted, too tired, too lazy, or was rushing to finish. Maybe all of them at once.
The tail design is not consistent in any of Asami’s Gura’s artwork which can be found in danbooru.
It’s hard for me to determine if these are full AI or traced because despite being able to zoom in, the images are very jpg artifacted and have a noise layer added on top.
I will update this section before publishing if I receive a reply.
Update: As of Nov 30 I never received a response to my email.
NEW ARTWORKS ALMOST EVERY SINGLE DAY
So I started this doc back in October, took a break, then resumed. There were roughly 48 artworks posted of her OC through 57 days (from October 3 to November 28), a new image posted almost every day. The account is basically all thirst bait art now but seeing how all of them have 20-60k likes and 100+ comments I’m sure the premium checkmark is going great for you, Asami.
On top of that, there’s the Vgen commissions that aren’t posted on twitter. There are 12 reviews from Oct-Nov, which means the total of artworks drawn in 57 days was over 60.
Is this art output normal? I would say no, but it’s not completely impossible. There are a few things to consider when it comes to art speed, such as canvas size and art style. An illustration with a background will take longer than a flat character on a white background, and a drawing done in a 800px canvas generally will take less time to finish than a drawing that’s 6000px tall.
However, after drawing a character so many times, you’d think there would be consistency in their anatomy proportions especially in two artworks drawn back to back.
October 22 October 24 (I made it worse by censoring it, sorry)
AI USERS DONT LEARN FROM MISTAKES
Going back to the timelapse, now that all of these new artworks were posted while I was away. The first thing I noticed across these new images is that she has not bothered to learn how to draw braids despite the character having a braided sweater.
The Oct 8 image is the one from the timelapse she had posted before. So we can see the thought process behind the linework for these.
The braids weren’t part of the sketch, so the lineart begins immediately. I slowed down the timelapse and turned it into two gifs so it’s easier to follow. There is only one instance where an area of the braids is redrawn or corrected, but the lineart is essentially still the same. There were no adjustments made.
Gif version→
October 12 October 13 October 18 October 21 November 1
If you keep tracing AI braids instead of learning how to draw them you are never going to draw them properly.
For those who actually care about learning, there are multiple ways and techniques to draw braids as there are many knot types, here’s a braid tutorial that I found useful when learning in the past.
No consistency between the sweater design.
What do AI braids look like?
Here’s an example 🡇
AI GENERATED AI GENERATED Asamiarts
For all of the reasons I have provided so far in this document, I believe that Asami’s skill jump from 2022 to 2025 is NOT genuine.
It’s not impossible, just highly unlikely. Please do not use improvement speed as the only reason to suspect someone of using AI.
There are real artists who practice and study hard and improve at an impressive speed in only 1-2 years.
My assumption is that this started sometime during 2023. I don’t have to wonder about the reasons or circumstances that would lead someone to build their career out of lies and fraud, it’s always money and/or popularity.
March 2023 vs August 2025
★ Request to Vgen
To respect commissioners and their OCs, I have avoided using art from Asami’s Vgen gallery but I want to talk about Vgen for a bit. There are multiple people tracers on their platform right now.
I understand that as the platform grows these fraudsters slip through the cracks, and I don’t know how the vetting process to join works right now (Last I heard it was invite codes). I don’t have time to make documents like this for every single tracer out there, but I had sent what I believe to be very substantial evidence when I reported some accounts last month. It’s been nearly two months (reports sent on Oct 8), yet action was never taken on the accounts. I never received a response, so I don’t even know if the reports went through. I was told that they get hundreds of reports so it will take some time to review them but in the meantime these people are still taking commissions from people who don’t know better. Please consider separating AI reports from general platform reports if you haven’t already as this is going to keep happening from now on.
AI gen has gotten extremely accurate, yet it’s not hard to catch if you ask for a PSD. But tracing AI is a more complex matter to investigate and creates distrust between all artists, especially beginners.
2025 AI Miku samples (Stable Diffusion)
AI AVATARS ON SOCIALS
Something I keep noticing is that accounts tracing AI sometimes have AI generated avatars. I am not sure if Asami generated these, as it's impossible to find a source online. There is also the possibility that the images were saved and used without knowing they were AI.
December 2022 Etsy avatar January 2023 Carrd website avatar
These are wayback machine snapshots, however the January 2023 avatar is still visible in her abandoned gumroad page.
This soulless eye style was fairly common back in 2022 when AI was not as advanced as it is today.
(Removed the part about hololive tracers)
The following are some previous cases of AI tracing
This is not a comprehensive list.
Tweet source / Document (if available)
IdoMia painting over AI Repost (18+ NSFW)
Haku Yuan tracing AI for commissions @haku_yuan and AI tracing/referencing/scamming
Juvvie tracing AI AI art/tracing allegations of VGen Artist Juvvie: A document supporting the User Report for AI/Tracing/Scam/Fraud
Previeww0 tracing and generating AI (tweet is SFW, Previeww0 account is NSFW)
Mothdusky tracing AI Mothdusky doc 2.0 The Biggest Fraud In Furry Art History (18+ NSFW)
Misouuy/Foame_ tracing AI @misouuy tracing over AI
Vonbabbitt tracing AI VonDoc2
As this becomes so prevalent I’m afraid that in the future people might wrongfully accuse others, whether in good or bad faith. Please don’t resort to hate or threats. Even if something might seem "very obvious" at first glance, please review these things privately first and ask other artists for a second opinion. The only reliable proof right now is live drawing in the same canvas (magma, drawpile, etc) with screen capture as timelapses and recordings can be forged.
12/5 Update
Asami’s response
I was not planning to update this document after releasing it, because I never intend to make these tracing cases an ongoing matter. I am only keeping this here for posteriority on how far someone is willing to go to pretend they are not tracing. Asami’s first response was a drawing calling this “drama”, followed by a tweet implying that I used her email to sign her up to NSFW websites (for the record, her work email has always been visible in her website so I didn’t leak it in this document) and that I am doing this because my drawings “don’t even get 1k likes” (lol)
Tweet source (Dec 3) / Tweet source (Dec 4)
This is not the first case of someone getting caught lying and immediately deflecting and accusing others of jealousy instead of providing evidence. The third response had an actual video recording of her progress. Before I get into that, let me point out something in her reply.
Asami is saying that when she opened her Etsy she used the first cute picture she found online as her profile image, but that was not the image that was used when her etsy was opened. Her Etsy was opened in 2021, I don’t know what month specifically but the first review is from December 2021. The AI profile picture appeared sometime during December 2022, so one year after her Etsy was created.
This was her profile picture back in November 2022, so the AI image was NOT her first profile picture. Note that in the doc I mentioned there was a chance it was used without knowing it was AI but I never claimed that AI was used for commissions back then.
More importantly, the video that she recorded and posted has been edited. I was expecting something like AR tracing, but given that she zooms in and out I ruled that out. What she did was chroma keying green out of the video during the sketching phase.
I can’t embed videos in Google Docs and it’s too long to make a gif so I will show screenshots of the relevant parts. Here is the tweet with the video, but I’ve also reuploaded the video here in case it gets deleted. (Update: Tweet was deleted on December 8)
1:16 7:09
Chroma keying is using color information to select a very specific range of colors and isolate them (source).
You might have heard the term green screen too, those who are familiar with OBS and streaming probably know about chroma key filters.
Green is completely absent from the color palette when she is sketching, after she’s done any instances of her opening the color palette show the green ranges properly. When she is zooming in and out the sketch, you can see the chroma key bleeding through.
These visual glitches only happen when she is sketching and completely stop after she’s ready to do the linework. If this was a video compression issue, it would also happen when she’s lining or coloring.
At 1:15 she uses the QuickMenu to make a new layer.
But at 2:02 you can see her quickly opening the layers tab for the first time in the video and Layer 1 is missing.
So the new layer she created with the QuickMenu at 1:15 for the lineart was Layer 3. If the video isn’t edited and the recording started right after she made a new canvas in Procreate, there would be no way for Layer 1 to have disappeared. It was deleted.
When you insert a photo in Procreate, that layer will be named Inserted Image. But if the first thing you do is insert a photo after making a new canvas, then that layer will be Layer 1.
The image below shows what it looks like if you don’t draw anything in Layer 1 after creating the canvas and instead insert a photo. The photo inserted becomes Layer 1.
I don’t care if you generate AI images. I don’t care if you want to trace AI instead of learning how to draw either, some people genuinely don’t mind and will still follow you. But if you are going to charge people money for it, you should be upfront about it instead of going this far to hide it.