In 1957 the newspaper Daily Express did a series of James Bond comic strips, one of the first visual depictions of the character (and likely the first design for the MI6 secret agent with double digit pictures), and one of the few ever to be not based on any actor’s likeness. Bond’s appearance in these strips was based on a private sketch commissioned and directed by Fleming himself, the design was then made less “outdated” and more masculine by the strip’s artist John McLusky. Until AI progresses to the point the original sketch alone is a sufficient dataset* I think this will be the nearest to Fleming’s vision of the character one can get.Prompt: DE007, black hair, asymmetrical hair, asymmetrical bangs, single bang, mature male (most of this wasn't specified in training in V1. Don't use V1) Not in data, but authentic to books: aqua eyes (“blue-grey”) Helps in some styles: combed hair Native style (use as negative for anything else): Comic, traditional media, monochrome, greyscale, hatching (texture), halftone, Negative: facial hair, source_anime (Facial hair is so strongly linked to adult male it needs to be negatived in a lot of styles, including realisitic. I find the character works best in un-anime styles) Potential Negative: speech bubble, blank speech bubble, thought bubble,curly hair, wavy hairFinding enough to assemble a dataset in good quality was surprisingly easy: A set of the original art went up for auction a few years ago, and the auction’s website had good quality scans for the stuff that was being sold. Taking these and a couple of promo pieces I found and cropping Bond out got me 54 pictures (Edit: 52 in V2. Two pics had really low quality hair in retrospect). Auto-tagger was surprisingly good here (I’m used to it being awful for non-anime), but not many good full body pics in TD. Epoch pics took forever to look anything like him, which I blame on the low tagging, very un-anime appearance and need for particular negatives (edit: V2 with better tagging did much better, on by Epoch 4).The scar Bond has in the books shows up a few times in the training data, but I only tagged it once ("scar"). Due to the artstyle most of the time rendering it as just a single straight line, even if you're looking for it you'll just mistake it for part of his face a lot of the time. The one instance that got it tagged was a close up that was unmistakably a scar (edit: they're trained in V2, but the results are awful). The wide mouth of this model is accurate to the source material, which puts the joke about the GoldenEye 007 boxart into a new prospective.*I wouldn’t be surprised if there was already some software that could convert it to a 3D bust that could then be used to make a dataset.

James Bond (Daily Express Comics)  / ジェームズ・ボンド (デイリー・エクスプレスのコミックス版)

Uses the dataset changes from Illustrious version.

PDV2_James_Bond_Daily_Express_Comics-000012.safetensors

In 1957 the newspaper Daily Express did a series of James Bond comic strips, one of the first visual depictions of the character (and likely the first design for the MI6 secret agent with double digit pictures), and one of the few ever to be not based on any actor’s likeness. Bond’s appearance in these strips was based on a private sketch commissioned and directed by Fleming himself, the design was then made less “outdated” and more masculine by the strip’s artist John McLusky. Until AI progresses to the point the original sketch alone is a sufficient dataset* I think this will be the nearest to Fleming’s vision of the character one can get.Prompt: DE007, black hair, asymmetrical hair, asymmetrical bangs, single bang, mature male (most of this wasn't specified in training in V1. Don't use V1) Not in data, but authentic to books: aqua eyes (“blue-grey”) Helps in some styles: combed hair Native style (use as negative for anything else): Comic, traditional media, monochrome, greyscale, hatching (texture), halftone, Negative: facial hair, source_anime (Facial hair is so strongly linked to ***** male it needs to be negatived in a lot of styles, including realisitic. I find the character works best in un-anime styles) Potential Negative: speech bubble, blank speech bubble, thought bubble,curly hair, wavy hairFinding enough to assemble a dataset in good quality was surprisingly easy: A set of the original art went up for auction a few years ago, and the auction’s website had good quality scans for the stuff that was being sold. Taking these and a couple of promo pieces I found and cropping Bond out got me 54 pictures (Edit: 52 in V1. Two pics had really low quality hair in retrospect). Auto-tagger was surprisingly good here (I’m used to it being awful for non-anime), but not many good full body pics in TD. Epoch pics took forever to look anything like him, which I blame on the low tagging, very un-anime appearance and need for particular negatives (edit: V2 with better tagging did much better, on by Epoch 4).The scar Bond has in the books shows up a few times in the training data, but I only tagged it once ("scar"). Due to the artstyle most of the time rendering it as just a single straight line, even if you're looking for it you'll just mistake it for part of his face a lot of the time. The one instance that got it tagged was a close up that was unmistakably a scar (edit: they're trained in V2, but the results are awful). The wide mouth of this model is accurate to the source material, which puts the joke about the GoldenEye 007 boxart into a new prospective.*I wouldn’t be surprised if there was already some software that could convert it to a 3D bust that could then be used to make a dataset.

&lt;p&gt;Uses the dataset changes from Illustrious version.&lt;/p&gt;

Files with SHA256: b238495fc80963bc845f21135d426f0af45801bd31bc5cfebec11c03c11cbee4