Advertisement
Guest User

Inafox to FA

a guest
Dec 14th, 2023
38
0
Never
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 14.13 KB | None | 0 0
  1. Ticket #236834 was closed without any objective determination/proof of how/why my art was regarded as AI.
  2. I will supply here various indisputable proof that show that I haven't used any AI. An AI can't explain rationalisations behind an image, but I can.
  3.  
  4. Originally the claim was the whole image I made was AI, now the goalposts have moved "yet again" that a completely different element of the image is "AI", and the element mention is only called a "part" of the image, not where or what even.
  5. To indicate a "part" of an image is AI and not say what is dubious, but I can try to do some guess-work to what you're mis-perceiving as AI and offer explanations. For example, non-AI artifacts, which come in two classes: anti-aliasing and blending.
  6.  
  7. Furthermore, whatever is considered to be "partially" AI in my image, I can re-draw or re-design, even timelapse, etc.
  8. Letting me know the element accused is important. That's why I find this verdict especially strange since artists are often asking how to get their art to look less AI, so much so there's blogs and media on it. But there's no definite answer from any one site or single moderator, just contentious vote systems. All this does is hurt artists, removing our art from the web, allowing actual AI posters to reap the benefit from our hard work. I don't think an AI user wouldn't put all the effort I am in defending my art, since they are scammers and are still allowed on many sites and they don't care. I'm probably in fact, spending more effort defending my art than I actually made it at this point. As you can see, this is very unfair and artists don't deserve that.
  9. https://parkablogs.com/content/how-prove-your-art-not-ai-generated
  10.  
  11. Original project file:
  12. https://1drv.ms/u/s!Aum-aQXf16OzbZv.....Tt-5w?e=x7kVQ0
  13. https://1drv.ms/u/s!Aum-aQXf16OzbBo.....pI_aY?e=dWkYDB
  14.  
  15. I can understand it's hard to tell if an image as AI but there's 3 clear main aspects that are core evidence that my artwork isn't AI:
  16. * The first is that the image stroke work and brush, etc, is consistent across the set of images created, and has handcrafted repetition and very simple soft rendering. And while I can't find a capture for the removed image, there are captures from other images in the group, to which I have not uploaded but possess the same motifs.
  17. * The second is that the rendered elements of the image possess liquify artifacts from radial antialiasing with an upward vector, AI doesn't do this, while my artwork here does, whereas a AI would actually "remove" such an indication and appear uniform like a photo taken instead of digital art:
  18. https://i.ibb.co/kc69Mhg/2.png
  19. https://fotoforensics.com/analysis.php?id=c4fa44d5988e19ea061821b14afec872dafae8cb.361592
  20. https://29a.ch/photo-forensics (alternative to the above as fotoforensics seems down now)
  21. * The third aspect is that if the coloured rendering is the accused element, then note that the rendering is trivial and more quickly and ironically more easily done with CSP tools than with AI and I can't imagine a scenario where someone would use AI to make a simple rendering that takes a few minutes tops, since AI images tend to take longer than that to process/prompt even due to their brute-forcey nature and heavy computation. You can find many much better renderings done by me in timelapses, so claiming a quickly made gradient with some lines and dots to be AI to me seems purely petty on the basis of "artistic merit". I will consider the shading separate from the rendering as it's linear and should be evaluated as such.
  22.  
  23. Since I don't know what elements are perceived as AI I'll go through some photo forensics and comparative proofing, that show my art clearly isn't AI.
  24. In my post I'll cover every possible aspect that may be mis-perceived as being AI, and highlight the clearly manual work.
  25.  
  26. Evaluations:
  27.  
  28. [ Colour consistency ]
  29. As in these examples, AI greyscale images tend to sample in blocks with lines, you can determine this by increasing the saturation of an AI image:
  30. https://ibb.co/dKfsXdg (google search "sketch by ai")
  31. https://ibb.co/rZRKR11
  32. While traditional art has more logical consistency of colour, to the line:
  33. https://ibb.co/yVZKcpM (google search "sketch art")
  34. https://ibb.co/yk613VN
  35. Evaluation: In the compressed version of my image, you can see the bleeding of the surrounding line colour where the brush has become contaminated, this is because my brush will pick up samples of surrounding colour or I will sample the line itself, as such there is more colour in the lines smoothly around the rendered area. Because my image is processed in 24-bit internally of Clip Studio, and CSP can only export 8-bit images, you can see this difference, while AIs today are only 8-bit, as such any obscured colours from an AI image when "saturated" will be far more "bitcrushed" than from real art. Since the compressed version of my artwork retains the colour, un-noisy, and the colour in the lines isn't simply removed, it's clear that my image contains colour-space impossible in AI images. The line work contains no AI colour artifacts, and the gradient layer is greyscale with a solid colour mapping.
  36.  
  37. [ Directional antialiasing of coloured areas ]
  38. It is evident when zooming in past 100% that my artwork possesses transformation and liquify artifacts, especially in the merged gradation layer:
  39. https://i.ibb.co/kc69Mhg/2.png
  40. https://i.ibb.co/WPCYczX/image-4366.png
  41. https://fotoforensics.com/analysis.php?id=c4fa44d5988e19ea061821b14afec872dafae8cb.361592
  42. https://29a.ch/photo-forensics (alternative to the above as fotoforensics seems down now)
  43. Evaluation: These elements are on layers, their directionality shows that not only were they painted on layers with soft brush, but with liquify adjustments in a specific heading. This is because originally the orifice was going to be open, and I prolonged the phallus. Since it's evidential that the image had been moved around consistently, any composition created by an AI initially would had been completely replaced, moved, destroyed, etc, to the point that the end result would be entirely human-made. I however have had no reason to use AI to make a gradient and specularity, wrinkles, etc, to which are clearly morphed and created entirely through CSP's tools with relative antialiasing. An AI image appears static in terms of ELA due to its diffusive properties, the presence of the high-amplitude red-blue shift show clear massive non-AI manipulations. AI is very good at smoothing out manipulations, humans are not.
  44.  
  45. [ Rendered edge ]
  46. It's clear that the rendering layers I had made are "leaky", meaning they go beyond the inner barrier of the line edge and even sit atop the base lines in my artwork to give a more traditional-esque edge gradient. AI images on the other hand would be proportional by flat area to the overall governed shape of the rendering. In other words, they'd have edges, too, and be proportional and not contain line leak to which only occurs when "expanding" the selective margin of the selection area.
  47. Evaluation: The edges of the paint work are clearly stroked, and not even masked by the lines themselves and have no "line" expected of an AI image. They are undulating brush work and leak over the edge, and there is more complexity in the clearly manual edge than the inner gradient as well, in fact a lot of the areas are very nebulously painted, you can also see where the brush line strokes at the top of the phallus are revealed in their contour and texture, prior their merging, showing indisputable brush layering.
  48. https://i.ibb.co/SRp924r/image-4360.png
  49. Some parts aren't merged as well:
  50. https://i.ibb.co/Cn1v2wk/image-4361.png
  51.  
  52. [ Misplaced white brush strokes ]
  53. Some parts of the sides of the rendered phallus is even translucent, not white, while in other areas the sides are painted translucently with white by accident, but as you can see in the top-left stroke were painted by the line brush and adheres the line art, and layered shown. I mention this because it's one issue that might have been misinterpreted as a photomanipulation:
  54. https://i.ibb.co/P9NYVfD/image-4371.png
  55. But as you can see, an AI can't possibly do this because it can't handle brush shape indistinguishable from the same-colour without having some form of edge like a line that is a consistent shape, it's very clear that the white strokes are on the outside of the coloured rendered area they are along, are brush convex strokes and are even accidentally placed in the facial area. All of these white brush strokes are isolated on unmerged layers.
  56. Evaluation: Misplaced strokes are the result of trying to fix edges crudely, this sometimes happens when an artist samples the background by accident instead of using the erase mode for the brush. Places of misplaced strokes include face, around the phallus, the left leg and the erased bed corner in the bottom right. My friend thinks this is one aspect that may be confused with photomanipulation, so pointing out these are obviously misplaced strokes I find is important to note and they only exist inside the group "Folder 1 4" to which was primarily made for the coloured parts of the image, nonetheless there are some both black and white strokes in it. Since neither the edges nor the inner-area is remotely indicative AI, it's clear the rendering does not resemble AI in any other possible terms and the only artifacts there are a product of low resolution antialiasing as regarded in the antialiasing evaluation.
  57.  
  58. [ Texture ]
  59. There is texture in the image which compresses unusually in the compressed version, I have noticed this and it's not present in the original project.
  60. If you make a selection from the "Stucco 1" layer and fill it, the original line work will be revealed, this is because I painted the sketch as a mask on a textured paper layer, as per CSP's limitations. You will notice that this shows far less noise than in the overall composition, as painted with a charcoal style smooth brush. You'll notice that this is also the same brush used on the rendered layers:
  61. https://i.ibb.co/VJ3SKjf/image-4359.png
  62. Evaluation: There are multiple passes to the sketch work, while some layers merged have irremovable texture, the texture usage is the same throughout and has the same overall resolution and tiled uniformity.
  63.  
  64. [ Line consistency to other works from the same set ]
  65. One of the trolls that did attack me saying my images were AI, they said stuff like "you wouldn't be able to reproduce that style in another drawing", while equally saying stuff like "your art is too inconsistent so must be AI". I've proven countless times my art isn't AI, showing my process, mixing mediums, etc. It's evident I'm being harassed and I'm not very happy that FA permits this mistreatment of artists of all kinds by just accepting and determining by hearsay and random decisions. I'm sure you understand this. It is whoever is making these false reports and leaving accusatory comments that should be reprimanded, especially considering they are pro-AI and are against FA's anti-AI policy, hence this whole artist-trolling thing they do.
  66. As mentioned, the artwork is part of a set of images I made, there are some GIFs I made that show me drawing in the exact same style. While not the same image, the linear quality is precisely the same, despite being the desktop version of CSP in these particular cases:
  67. https://i.ibb.co/4M4rMMC/1.gif
  68. https://i.ibb.co/JFMgWyG/3.gif
  69. https://i.ibb.co/WnjTyXJ/2.gif
  70. https://i.ibb.co/SyHjf0F/4.gif
  71. https://i.ibb.co/GFKJQQd/5.gif
  72. https://i.ibb.co/18qdBzW/6.gif
  73. The last GIF, from the same set, doesn't just involve me drawing the contour once but several times, as it is a sticker, like in my sticker art from before.
  74. Evaluation: The line work and brush has the same tapers, density flow, colour noise, tuft density, etc, as in the uploaded image. The work here and in the removed is relatively low quality compared to my artistic abilities, though some trolls claimed that these images are "below" my level. There are timelapses of me painting, doing 3D work, etc, on my social media feeds. Such claims are bogus. Industrial artists work on various things and change their style to suit the project, I likewise like to "change" my style up regularly. Most of my art isn't furry either and coming to/from furry always introduces new skillsets carried over from my more studious artwork, furry art is just a hobby for me. Throughout the image set there are various elements of consistency and inconsistency to which AI is simply incapable of, much less is it capable of understanding the overall style here, and is persistent throughout all my art in for example, the forepaws:
  75. https://i.ibb.co/DDpbgxH/image-4364.png (not the same character, yet same design near enough, same design style in another position, same tufting)
  76. https://i.ibb.co/m5ztTfN/image-4365.png (the character from the GIFs)
  77. The art in this latest inking set is "tuftier" than a lot of my other art, as I tend to avoid tufts for painterly art to which I usually do as I find tufts annoying to paint but not draw. But otherwise doesn't differentiate to my higher-res work.
  78.  
  79. I find it important to remember that the rules the community elected for against AI are to:
  80. * Protect against plagiarism, since AI steals art
  81. * Protect human merit and effort, since AI steals labour
  82. * Protect hard-working artists, since AI steals livelihoods
  83. * Protect FA's PR in terms of being a safe site for artists, since genuine artists need safe spaces
  84.  
  85. Future proofing possibilities?
  86. * Returning to vector layers (which is annoying because CSP broke them so I might need other software now and can't afford atm)
  87. * ... any other ideas?
  88.  
  89. I hope this is enough evidence to consider. It may also be helpful in future for detecting actual AI images as others have mentioned they've seen quite a lot of images on here that are clearly AI and garnering undue attention.
  90. I also hope you can be more impartial and treat your users more equally, as most people aren't asked for project files and WIPs, etc. It's discriminatory, and as someone who has received transphobia, misogyny, interphobia, ableism, and other aspects from the same trolls on Discord, etc, I can only take this as bullying and harassment from these trolls. So please don't single me out and be mindful of the overall situation I've been experiencing.
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement