Show HN: Create Comics Using Stable Diffusion | Hacker News Reader

Show HN: Create Comics Using Stable Diffusion

comicsmaker.ai

59 pointscom-adm24 comments

How does this create consistent characters between frames?

My understanding was that this is very hard to do reliably without dreambooth style fine tuning, which is expensive and slow

Simplest way is use celebrities, mix them together, switch their race/gender etc. More discussion here: https://www.reddit.com/r/StableDiffusion/comments/12djxvz/ho...

Link to the relevant comment..

https://www.reddit.com/r/StableDiffusion/comments/zv5mbq/sim...

Nice concept, just don't expect to have any copyright to the (images in the) comic, as the USPTO has mentioned [0]. If you used an AI text generator as well, expect to not have copyright to that either. Not that that's necessarily a bad thing, copyrights and patents are inherently anticompetitive.

https://www.theverge.com/2023/2/22/23611278/midjourney-ai-co...

What if the human author considers the AI as a tool and does not credit the AI in any way?

There's the rub, isn't it? At a certain point of AI progression, if you don't point it out, no one would know it's AI, as it's indistinguishable from a human creation. Even tools based on AI to detect AI creations will themselves succumb to the fact that the generator AI can use the information gleaned by the discriminator AI to improve; this is in fact how generative adversarial networks work.

I wonder if soon in the future highly skilled humans will have to prove their works weren't made by AI.

> If the Office becomes aware that information essential to its evaluation of registrability “has been omitted entirely from the application or is questionable,” it may take steps to cancel the registration. 44 Separately, a court may disregard a registration in an infringement action pursuant to section 411(b) of the Copyright Act if it concludes that the applicant knowingly provided the Office with inaccurate information, and the accurate information would have resulted in the refusal of the registration.

From https://copyright.gov/ai/ai_policy_guidance.pdf

Is the distinction between generating full works via AI and using AI in a manor indistinguishable from Grammarly respected when admitting to any use of AI?

Lengthy copyright may be uncompetitive, but the absence of it would also be surely? If I create something and an already large business can just take it and use it without any recompense, doesn't that merely entrench their market position?

Can you explain your reasoning?

>If I create something and an already large business can just take it and use it without any recompense, doesn't that merely entrench their market position?

Isn't that the entire basis of OpenAI and GPT?

> If I create something and an already large business can just take it and use it without any recompense, doesn't that merely entrench their market position?

There's already numerous posts from artists claiming their work was copied verbatim (or near-verbatim) by major corps. Fighting it in court will take years and cost a fortune they don't have.

That may be, but is the solution to get rid of what little protection they legally have?

If it's not protecting them, and it's causing other harms?

You would have to demonstrate that the harms far outweigh the goods for all works protected by copyright, not just the poor artists you mention.

Law is never perfect. I agree if it does more harm than good for most then it should be replaced.

You could always retouch the images in photoshop. I believe that ruling was for fully generated AI, with no retouch afterwards.

That interpretation probably does not pass muster. The guidance from the US Copyright Office leans on substantial human authorship or creative changes or interpretation of the output of AI generative tools.

Probably better to just link the actual guidance, it’s only a couple pages.

https://copyright.gov/ai/ai_policy_guidance.pdf

From the linked document. To me this is the same interpretation, but IANAL.

Individuals who use AI technology in creating a work may claim copyright protection for their own contributions to that work. They must use the Standard Application, 39 and in it identify the author(s) and provide a brief statement in the “Author Created” field that describes the authorship that was contributed by a human. For example, an applicant who incorporates AI-generated text into a larger textual work should claim the portions of the textual work that is human-authored. And an applicant who creatively arranges the human and non-human content within a work should fill out the “Author Created” field to claim: “Selection, coordination, and arrangement of [describe human- authored content] created by the author and [describe AI content] generated by artificial intelligence.”

In particular, this sentence:

> “Selection, coordination, and arrangement of [describe human- authored content] created by the author and [describe AI content] generated by artificial intelligence.”

Sounds to me like a DJ selection or like Duchamp's found objects as the creative element.

How does it handle having to generate the same character in multiple consecutive panels?

Can I define that character once and refer to them later, e.g. “John is looking at his phone”?

Simplest way is to use celebrities. You can mix them like "HenryCavillKeanuReeves" "KeanuReevesEmmaWatson" switch races/gender etc. since the model's training data contains lots of celebrities, it would be able to produce consistent images

in "Create Characters" page you can save characters(eg: jack), and later in the Prompt box in "Panel Creator" you can refer the characters with @ mention (eg: @jack looking at a phone) internally the prompt will be updated with the sub prompt for the character

I already made an assignment for my classes next semester (before seeing this) to create a graphic novel with images generated in Bing's Image Creator (http://Bing.com/Images/Create) They will generate images, then use Photoshop or another image editing tool to fix faces, hands, and other details.

Mobile now so can't play with it Looks interesting. Does this only support manga or are Bande dessinée artists styles also possible?

should work on Mobile, but desktop would provide best experience. Other than Manga, it supports DC/Marvel style which uses a model trained specifically on western comics, but there is always the "General" style which is the base stable diffusion model which can create any style with a bit of prompt engineering

Nice work! I had actually considered this idea before, but had trouble figuring out how to keep the character designs consistent.

Reply on news.ycombinator.com