Asgard - Stargate SG-1

The grey aliens from Stargate SG-1 - Thor and his people. Trained on 68 captioned stills. Trigger word: asgard. Use at strength 1.0.

As far as I can tell this is the first Asgard LoRA for Krea 2. There is a good one for SD 1.5, but nothing on this base model, and Krea 2 on its own turns "asgard" into a generic grey alien with the wrong proportions.

What it brings

Dome skull, black almond eyes, thin grey limbs, and the correct head-to-body ratio. That last one is what the base model gets wrong.

Close-ups work. This is the main reason the LoRA exists. Without it an Asgard face falls apart at close range and you have to keep them small, distant or in silhouette. With it you can go right up to the eyes.

SGC uniforms with the patch and the cold series lighting come along as a bonus. They were in the source material and bound themselves to the trigger. Composition stays free - a wide shot stays a wide shot, the LoRA does not drag everything toward a portrait.

One known deviation: body height lands around 1.20 m instead of the canonical 1.00 m. For illustration work this has never bothered me.

Two limits you cannot prompt your way around

Neither of these is Asgard-specific. They are limits of the text encoder, and I only mention them because they cost me a lot of wasted generations.

Size relations do not work. I tested three phrasings of the same instruction on an identical seed: "towering over the asgard by more than twice its height", "the asgard's head reaches only to the soldier's hip", and "top of the head is level with the belt". All three produced the same factor of about 1.5. Comparatives do not get through, and neither do absolute anchors. The way around it is composition, not language: put the figure alone in frame next to an object of known size (a console, a doorway, a chair) and set the camera low. The model scales to the object. Two images in this gallery do exactly that - the corridor shot and the kneeling soldier - and those are the only ones where the scale reads correctly.

Counts are ignored. "seven tall seats, five holding an asgard, two empty" gives you five to nine seats in varying occupancy. If the number matters, it has to happen in post.

Prompting that works

Describe the scene, not the creature. The trigger carries the anatomy on its own. Proper nouns from the show carry real weight: Orilla, Asgard High Council, SGC all land.

One thing worth knowing: wide council-chamber shots are the hardest thing to get right, because they ask the model for counting, arrangement and anatomy all at once on a small area. Those are exactly its three weakest skills. Moving the camera closer solves more than changing the seed does. The strongest images in this gallery are the ones with a human in frame - the contrast gives the base model something familiar to build the room from, while the LoRA only has to carry the figure.

Please note

Fan project, not affiliated with or endorsed by MGM. Unlike a character LoRA this one depicts no real person, it is a creature design.

Some images in this gallery also use "Stargate and DHD (SG-1)" by lamarsha (civitai.com/models/2777074) for the gate itself. Credit where it is due - the gate is lamarsha's work, not mine.