WotC: 'We made a mistake when we said an image not AI'

It seems like AI art is going to be a recurring news theme this year. While this is Magic: the Gathering news rather than D&D or TTRPG news, WotC and AI art has been a hot topic a few times recently.

When MtG community members observed that a promotional image looked like it was made with AI, WotC denied that was the case, saying in a now-deleted tweet "We understand confusion by fans given the style being different than card art, but we stand by our previous statement. This art was created by humans and not AI."

However, they have just reversed their position and admitted that the art was, indeed, made with the help of AI tools.

Well, we made a mistake earlier when we said that a marketing image we posted was not created using AI. Read on for more.

As you, our diligent community pointed out, it looks like some AI components that are now popping up in industry standard tools like Photoshop crept into our marketing creative, even if a human did the work to create the overall image.

While the art came from a vendor, it’s on us to make sure that we are living up to our promise to support the amazing human ingenuity that makes Magic great.

We already made clear that we require artists, writers, and creatives contributing to the Magic TCG to refrain from using AI generative tools to create final Magic products.

Now we’re evaluating how we work with vendors on creative beyond our products – like these marketing images – to make sure that we are living up to those values.

This comes shortly after a different controversy when a YouTube accused them (falsely in this case) of using AI on a D&D promotional image, after which WotC reiterated that "We require artists, writers, and creatives contributing to the D&D TTRPG to refrain from using AI generative tools to create final D&D products."

The AI art tool Midjourney is being sued in California right now by three Magic: The Gathering artists who determined that theirs and nearly 6,000 other artists' work had been scraped without permission. That case is ongoing.

Various tools and online platforms are now incorporating AI into their processes. AI options are appearing on stock art sites like Shutterstock, and creative design platforms like Canva are now offering AI. Moreover, tools within applications like Photoshop are starting to draw on AI, with the software intelligently filling spaces where objects are removed and so on. As time goes on, AI is going to creep into more and more of the creative processes used by artists, writers, and video-makers.

 

[trappedslider]

We're speaking of an artist training a model to create intermediate layers so he can pass off AI-assisted work for non-AI-assisted work in order to breach a contract. WotC suppliers would need to make a lot of money to make it viable, even if, as you rightly pointed out, the price of training data is dropping. I don't see anyone doing that honestly in the near future.

The only price that i'm aware of is the cost of a GPU depending on what you're going for (Well also electric bills)

 

[Jfdlsjfd]

The only price that i'm aware of is the cost of a GPU depending on what you're going for (Well also electric bills)

Indeed, but training a model from scratch require thousands of A100-day equivalents. I can't see a model being trained totally for that on a consumer grade GPU. And I am not certain (but lack the techical knowledge to be affirmative) that a LORA could suffice for providing convincing work-in-progress.

 

[Sorcerers Apprentice]

To make such a model you'd also need a lot of training data, ie. load and loads of sequences of WIP pictures. Artists sometimes post WIPs, but they rarely share the entire edit history, so getting enough data to train a model would be challenging, to say the least.

You can probably make a LORA to generate pictures that look like WIPs, but generating a convincing sequence of WIPs where it looks like each one is progressing from the previous one is going to require an entirely new model.

 

[Yaarel]

It will be impossible to determine in short order. Again, I remember the 'art' we generated via these tools over a year ago. It was a hallucinatory mess of color, and while interesting, was not all that impressive.

Fast forward to today, and we could already improve of a vast amount of WotC/D&D art with AI generated pieces.

Fast forward another year? Good luck telling anything apart.

Yeah. At this moment, AI art is at best a tool for artists. The AI generation requires completely a human to select the various outputs and then to extensively massage and finetune the results.

But in some years to come, excellent AI images can be entirely autonomous, and often indistinguishable from human efforts.

 

[trappedslider]

Indeed, but training a model from scratch require thousands of A100-day equivalents. I can't see a model being trained totally for that on a consumer grade GPU. And I am not certain (but lack the techical knowledge to be affirmative) that a LORA could suffice for providing convincing work-in-progress.

To make such a model you'd also need a lot of training data, ie. load and loads of sequences of WIP pictures. Artists sometimes post WIPs, but they rarely share the entire edit history, so getting enough data to train a model would be challenging, to say the least.

You can probably make a LORA to generate pictures that look like WIPs, but generating a convincing sequence of WIPs where it looks like each one is progressing from the previous one is going to require an entirely new model.

It took me a couple of days but I realized something, you want to treat the way someone uses AI generation as if it was the same as other art styles when it's not.

The normal way "art" is made with AI is you start with a first image's prompts, then once that's made you can further tweak it by doing image to image then some Inpainting and lastly upscaling.
before I did inpainting I did a single image generation others more than likely would do more

Spoiler

Spoiler

After inpainting

Spoiler

Upscaling

Spoiler

The guy who won the Colorado state fair went though a number of steps before picking his choice for entry. PNG info

Spoiler

parameters
annakendrick as a human rogue wearing leather armor on a rooftop
Negative prompt: deformed, bad anatomy, disfigured, poorly drawn face, mutation, mutated, extra limb, ugly, disgusting, poorly drawn hands, missing limb, floating limbs, disconnected limbs, malformed hands, blurry, ((((mutated hands and fingers)))), watermark, watermarked, oversaturated, censored, distorted hands, amputation, missing hands, obese, doubled face, double hands,canvas frame, cartoon, 3d, ((disfigured)), ((bad art)), ((deformed)),((extra limbs)),((close up)),((b&w)), wierd colors, blurry, (((duplicate))), ((morbid)), ((mutilated)), [out of frame], extra fingers, mutated hands, ((poorly drawn hands)), ((poorly drawn face)), (((mutation))), (((deformed))), ((ugly)), blurry, ((bad anatomy)), (((bad proportions))), ((extra limbs)), cloned face, (((disfigured))), out of frame, ugly, extra limbs, (bad anatomy), gross proportions, (malformed limbs), ((missing arms)), ((missing legs)), (((extra arms))), (((extra legs))), mutated hands, (fused fingers), (too many fingers), (((long neck))), Photoshop, video game, ugly, tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, mutation, mutated, extra limbs, extra legs, extra arms, disfigured, deformed, cross-eye, body out of frame, blurry, bad art, bad anatomy, 3d render
Steps: 50, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 2486026595, Size: 512x768, Model hash: c15090d329, Model: dreamscapesDragonfireNEWV20_dsDv20, Denoising strength: 0.75, Mask blur: 4, Version: 1.6.0

I took several steps when I made my custom MTG sleeves using AI

 

[Sorcerers Apprentice]

@trappedslider
The discussion is about a potential model to create a "fake" edit history to make it possible to pass off an AI work as if it was hand made, for fraudulent purposes. So it would have to mimic the traditional art process, not the AI art one.

 

[Jfdlsjfd]

It took me a couple of days but I realized something, you want to treat the way someone uses AI generation as if it was the same as other art styles when it's not.

We were talking in the context of how an artists could bypass a test made by WotC (who doesn't want AI art) to ask for proof that the art was handmade, not AI-assisted. Generating a model that create work in progress like if it was started from a hand drawn sketch is a possibility, but would require a dedicated model. Taking a picture of the artists while standing in front of the canvas would be, but on the other hand I have a photo of me saving the earth by shattering an asteroid with my fists, wearing the Golden Saint armor of Sagittarius, so... The only way I can imagine is actually sending the physical material during the creative process, but it's quite complicated. Asking for regular update of the digital files at several steps might work.

I think we misunderstood each other about the intent!

(exquisite picture btw, while it has the tell tall signs of AI, it's very nice and good enough for general use).

 

[trappedslider]

@trappedslider
The discussion is about a potential model to create a "fake" edit history to make it possible to pass off an AI work as if it was hand made, for fraudulent purposes. So it would have to mimic the traditional art process, not the AI art one.

We were talking in the context of how an artists could bypass a test made by WotC (who doesn't want AI art) to ask for proof that the art was handmade, not AI-assisted. Generating a model that create work in progress like if it was started from a hand drawn sketch is a possibility, but would require a dedicated model. Taking a picture of the artists while standing in front of the canvas would be, but on the other hand I have a photo of me saving the earth by shattering an asteroid with my fists, wearing the Golden Saint armor of Sagittarius, so... The only way I can imagine is actually sending the physical material during the creative process, but it's quite complicated. Asking for regular update of the digital files at several steps might work.

I think we misunderstood each other about the intent!

(exquisite picture btw, while it has the tell tall signs of AI, it's very nice and good enough for general use).

I honestly doubt someone would go to the trouble to do that...humans are after a lazy group in general

 

[Sorcerers Apprentice]

I honestly doubt someone would go to the trouble to do that...humans are after a lazy group in general

That's the point, for any practical purposes it can't be done, so demanding WIP files from artists is and will continue to be a safe way to ensure you're not paying for AI generated art.

 

[Jfdlsjfd]

Just to illustrate what I meant by the state of open, fully public domain/ open licensed models, here are some results, which are a throw in the past by a few months compared to what other models can do.

The result are raw unmodified generation, really "low effort", with extremely basic prompting and taking a few seconds to generate, no specific tool and no workflow to speak of. And yet even in this experimental stage, I wouldn't cringe at seeing them in a RPG product I buy (for the little time I spend looking at pictures anyway).

An eerie castle in a forest clearing at night, ominous atmosphere, a green moon in the sky.

An abstract painting in the style of Mondrian

a cloud giant

Obviously, people who tried harder than me got better results even in this preliminary state.

Between this and Adobe's Firefly, ethical models will soon be widely available. The unclear copyright status of scraping the Internet for dataset will probably never to have to be resolved if better-trained base models arise before the courts and legislators clarify it.