Lovecraft in 3D + Stable Diffusion. Made by taking the Meshbox 3D Lovecraft poseable figure (purchase | free face expression presets) for the Poser software, improving the skin a bit by tinkering with materials, and then doing a quick simple render in Poser with a 50mm virtual camera (a 50mm lens is best for portraits).

Then I used this quickie render from Poser in a ControlNet (simple ‘canny’ and ‘depth’) for AI image generation. Which in InvokeAI 3.0 and Stable Diffusion 1.5 produced a photoreal mugshot…

Then slight Photoshopping to fix the hair (little duck-tail quiff: ‘AI say no’), desaturate the skin a little, and vignette. And… the proof-of-concept portrait was done. It’s a bit ‘waxworks figure from del Toro’s basement’ if you look at it for too long. Fish-eyes too (he’s going to Innsmouth soon). But it’s reasonable for a first test without any work on the expression or pose.

More new additions to Stable Diffusion 1.5 LORAs, used for text-to-image AI image generation. Pulp / sci-fi ones such as these, of possible interest to Tentaclii readers, are swamped by the zillions of LORAs being produced daily for each-and-every Japanese anime character. But they do appear quite frequently, for those willing to hunt for and find them among all the new fluff.

Art Deco Architecture – v1.0. Of obvious use for 1930s and 40 sci-fi backdrops. Seemingly trained on wide-scale deco concept art produced by architectural illustrators of the period.

Sumerian Architecture – v1.0.

One can combine LORAs for generating the same image, so the combinatorial possibilities should be obvious for science-fantasy illustration. Possibly these two may combine in interesting ways with the also-new Multiversal Vistas – v1.0, said to be trained on “strange futuristic vistas, forgotten technology and monumental artifacts”.

Also of possible interest, though this time to the retro crowd, is the 1950s style but highly polished Romance Comics – v1.0. (Warning: can also do nudity). Seemingly trained on slick cleaned reprints, rather than scans of the original yellowing comics. SD 1.5 knows what Lovecraft looked like, so I guess one might be able to have this produce a comic featuring Lovecraft. With the era-correct suit, shirt, shoes. But I’ve yet to try that.

(Limbo) Liminal Space Style – v1.0. In the influential ‘spooky silhouette’ style of the ground-breaking Limbo videogame.

And Dictionnaire Infernal (updated today).

New free SD 1.5 LORAs of possible interest to pulpsters, RPG-ers, self publishers, and others. These have appeared this week on CivitAI, and are for use with free desktop AI image generation software such as InvokeAI.

Cross-section Underground

Style of Roy G. Krenkel (Golden Age comics artist and illustrator)

Style of Jean-Pierre Gibrat – v2.0 (lesser-known French BD artist)

TangoOne – Gil Elvgren (retro U.S. pin-up style). Specifically for use as an add-on to the M4RV3LS & DUNGEONS model.

Eric Powell Style (comic artist, looks suitable for depicting Lovecraft’s Innsmouth).

Hurrah! A very kind benefactor, who was upgrading to a blisteringly fast 40x series graphics-card, gifted me his old GeForce RTX 3060 12Gb graphics-card. This worth-$250 hunk o’ joy has been slotted in to my PC and suitably wrangled. Thus I’m now a proper local/desktop AI image generator. Since I also managed to install InvokeAI 3 on Windows 7 (it runs Stable Diffusion models), with a little .DLL swop-out trick, and everything AI is working fine and fast with the new card.

Nicely timed, released today… the free Alienscape – Strange Landscapes LoRA for SD 1.5. Generates the sort of landscapes you might have seen on an old SF paperback cover, from your descriptive text prompt.

The other thing I’m enjoying is the Stable Diffusion community ethos, once you break free from the paywalled online service-providers. Everything is free, once you have the card to run it locally. UI’s, models, add-ons, tutorials, workflows. All free. Nice.