Donmai

Deleted pixiv posts

Posted under General

Fenton said:

You need to put the original source link regardless of the location where you found the image. You may need to be careful, because some users in Gelbooru also upload paid rewards and put the original source link on their posts.

All I can find is third-party uploads. I guess there's no way to know if it's safe or not now.

Nameless_Contributor said:

If it was freely available on the artist's Pixiv it is fine to upload.

That's cool, I didn't know that.

Samsara_Kama said:

I like to add the archived commentary and mention that the source was deleted, even if I can confirm that it's the original source of the image (post #7611237).

Where can you find archived commentary?

gzb said:

Where can you find archived commentary?

You can find commentary titles using SauceNAO or ascii2d. The Wayback Machine seems useless for Pixiv artwork pages, but you can also find the title in the tab name.

Samsara_Kama said:

The Wayback Machine seems useless for Pixiv artwork pages, but you can also find the title in the tab name.

Pixiv apparently refuses to let their pages be saved onto the Wayback Machine, hence why it's nigh impossible to find archived sources from there.

help:image_source says if you find it on the internet, paste the URL, and danbooru will figure the source out.

If you want to upload an image from the internet (such as from Pixiv, Twitter, etc) you can just copypaste the link onto the upload page and the site will automatically set the correct source.

For gelbooru images, danbooru uses gelbooru in the source field.

I changed the image source accordingly.

reg_panda said:

help:image_source says if you find it on the internet, paste the URL, and danbooru will figure the source out.

For gelbooru images, danbooru uses gelbooru in the source field.

I changed the image source accordingly.

In cases where you are 100% certain the file MD5 matches the one that would've been retrieved from the now-dead source, it's fine to set the source to this dead link. Some boorus do modify metadata, but many do not (just like danbooru), so in those cases it's fine to change the source to the would-have-been source.

Yeah, that makes sense in theory (it still needs to be written in the wiki tho).

In practice that never happens, as SauceNAO don't say anything about md5, metadata or pixel hash equivalence. At least for unlogged users. It only says the picture is 97.66% the same. (AFAIK it is sauceNAO's way to tell 100%, still not very convincing as "100% certain" for "MD5 matches", rather the opposite.) https://saucenao.com/search.php?url=https://cdn.donmai.us/original/81/74/8174c73ec1bc1fcbb9bd40d289d3048b.jpg

Updated

reg_panda said:

Yeah, that makes sense in theory (it still needs to be written in the wiki tho).

In practice that never happens, as SauceNAO don't say anything about md5, metadata or pixel hash equivalence. At least for unlogged users. It only says the picture is 97.66% the same. (AFAIK it is sauceNAO's way to tell 100%, still not very convincing as "100% certain" for "MD5 matches", rather the opposite.) https://saucenao.com/search.php?url=https://cdn.donmai.us/original/81/74/8174c73ec1bc1fcbb9bd40d289d3048b.jpg

My comment wasn't referring specifically to matches found with saucenao, just in general.

As far as I'm aware Gelbooru is not one of the boorus that fucks with image metadata, so in that case it would be fine to change the source field to the correct Pixiv URL.

That's even worse, trusting someone elses word about the source. Trusting SauceNao code is one thing, but a random claim about the source is another. Why not just .. write your source in the source field? help:image_source definitely doesn't say that you should believe random claims about sources on the internet, and write that in the source field.

Also boorus and their users lie about sources, and modify the images all the time. Malice, mistakes, incompetence, and such. For example in this thread it is straight up recommended to manufacture a fake source. (Which may be the correct source, even may be provably the correct source. I think SauceNao gives more information for registered users about their data, and the matches.)

I don't see much advantage in copying source fields from other boorus. A disadvantage is that it strengthens misinformation.

As far as I'm aware Gelbooru is not one of the boorus that fucks with image metadata, so in that case it would be fine to change the source field to the correct Pixiv URL.

Keep in mind that gelbooru doesn't download the images from other sites as danbooru does. Also it doesn't fill the source field automatically. Both are done manually by the users. You don't trust "gelbooru" when you believe the source field, but the user that uploaded the picture. The user can modify the picture (crop, change colors, strip metadata etc), or accidentally modify the picture (resize, convert). Also can lie about the source, or miscopy it (when they upload a lot of images).

reg_panda said:

Keep in mind that gelbooru doesn't download the images from other sites as danbooru does. Also it doesn't fill the source field automatically. Both are done manually by the users. You don't trust "gelbooru" when you believe the source field, but the user that uploaded the picture. The user can modify the picture (crop, change colors, strip metadata etc), or accidentally modify the picture (resize, convert). Also can lie about the source, or miscopy it (when they upload a lot of images).

As a Gelbooru moderator, I can confirm this is true. Believe it or not, 98% of the time, our users are able to download the original image and upload it without issue. However, without checking the original source, you can never be sure. I've had sneaky little bastards try to hide their uploads of paid rewards by sourcing the smaller/censored Pixiv post, or crop out a character they didn't like while sourcing the unedited Pixiv post, or edit a dark-skinned male's skin color while sourcing the original color Pixiv post and so on. Some power uploaders you can be pretty sure about, but because we don't automatically download the image for the uploader, it's always a gamble.

ascii2d does list file md5 for results.
For the post in question here: https://ascii2d.net/search/color/8174c73ec1bc1fcbb9bd40d289d3048b
You can see the md5 on our post does match what the pixiv post had when ascii2d crawled it, so I'm going to change the source back to that.

In the absence of such records, though, I wouldn't trust any third-party source. In my experience on the 'net, Danbooru is an extreme outlier in its diligence with sourcing and quality -- technically and community-wise -- and even we have plenty of fuckups from inexperienced users.
I also wouldn't go off a similarity percentage from a reverse search (even if it did say 100%!) because AFAIK those are all based on models that almost never have a concept of pixel-perfection.

IMO, a dead first-party source is better than a live third-party one if we can be totally sure it matched when it was alive, so if we can verify via ascii2d or some other sort of archive we should use the original source.
But a misleading source is worse than none at all, because we shouldn't assert that we have the original (discouraging further search for it) when we might not.

1