More looking through CivitAI, which is the main repository for free generative AI image models and plugins. There’s a lot there, and it can take a lot of digging to find stuff.

There’s some obvious Cthulhu / Giger / pulp etc stuff. But here are some other interesting non-obvious items, which seem to have Lovecraftian and RPG potential, and which you would never find via search.

These two might be combined in interesting ways… SAMO bones carving (intricate ivory carving) and Occult Geometry.

You would be able to use negative prompting to steer the Bone Carving away from an Anime / Chinese look, and more toward Giger / Lovecraft.

And for RPGs and illustrations involving museums, explorers etc…

KnollingCase (puts X inside an old fashioned specimen case for museums).

Antixdisplay (puts X inside a modern museum display case).

And for Victorian / Edwardian explorers, Muttonchops / Sideburns (Realistic) (for men) and DrawingSD1.5 (19th century pen and wash pictures) for the explorer’s old books of field-notes.

How far AI image generation has come, in just a year. What will another year bring?

A new ‘fannish-but scholarly’ site of possible interest to Tentaclii readers, Comic Book Historians.

More delving into the options for ‘add-ons’ that can steer the Stable Diffusion 1.5 AI image generator toward certain styles or objects in its images. The community’s work on SD 1.5 add-ons now has an excellent range and depth (things move fast in AI-land), and also a healthy dash of ‘quirky’. The ethos is ‘free’, and there’s no payment involved so long as you run the image generation on your own PC. Image generation can be helped along by ‘guide poses’ created from 3D-figure desktop software such as DAZ Studio and Poser.

For the style of the French comics artist Moebius with SD 1.5 I found the following, after a thorough search:

* Moebius Color Style – v1.0 is a LORA that tries to emulate the later ‘colour Moebius’ style. Said to work best as an add-on for the large Xenogasm checkpoint model for SD 1.5, which seems to have been specially trained for the sort of subject matter you might find in old 1980s copies of comics magazines such as Heavy Metal, 1984, Epic, etc and similar. Perhaps also Zona 84, Ere Comprimee etc, at a guess. The 1970s-style Xenogasm model is NSWF in the eyes of the 2020s, as you might expect. Apparently Xenogasm 2.1 is the last before it started being heavily re-worked for more realism, so you may want to try that first.

* WASMoebius V1, which is an Textual Inversion ‘Embedding’ add-on for SD 1.5 rather than a LORA add-on.

* For a more old-school ‘B&W Moebius’ one might try the Centerflex – v2.8 checkpoint model. Its main promo images are photoreal and it is a “photorealistic-first” model. But delve into the description and note that this is a dual model. Switch it to its illustration mode and it is said to specialise in the ‘clear ligne’ line-art of the French and Belgian BD comics and Moebius in particular.

Artist styles that this model was explicitly trained to on include Roy Krenkel, Jean-Pierre Gibrat, Milo Manara, Willy Vendersteen, Francois Schuiten, Edgar P Jacobs, Herge, Jacques Tardi, and Moebius. […] This model contains a general activation keyword phrase to evoke a certain hybrid comic book style and composition. This phrase is “ligne claire” meaning “clear line” after the Franco-Belgian comic book [line-art] tradition.

* Possibly also worth a look, in complete SD checkpoint models, is the Protogen Infinity checkpoint model. Trained on comics, isometric art, science-fiction. As well as Protogen, ReV Animated (aka ReVAnimated) is a model name that pops up for use with the more comics-oriented LORAs. There are also SD 1.5 models trained on modern western superhero art, such as iCoMix.

* Metal Hurlant Comics – Moebius, Bilal, Druillet is a LORA trained on a range of the classic 1970s/80s French comics artists. Though it appears to work best with the more grungy Druillet style. An ‘alpha release’, and not likely to be updated to a 1.0 release.

Another Damn Art Model (ADAM) was partly trained on Moebius, along with other Heavy Metal and 1970s artists.

* There’s also a relatively refined attempt at Enki Billal – v3.0, a Bilal style which might combine well with one or two of the above. Update: And another attempt, Enki Bilal Style – v1.0.

Some people combine checkpoint models and LORAs and ’embeddings’. So for instance one might combine the big Centerflex model with the smaller Moebius Color Style LORA as an add-on.

I don’t see anything that can do the distinctive ‘dash shading’ of Moebius. Although I’m not sure an AI could do this, unless it could be trained to follow the shading on a depth-map from a 3D figure / scene…

Ok, that’s what I found. Now all I need is the 12Gb graphics card for the PC, to be able to try them out. 🙂