Verified: a year ago
SafeTensor
The FLUX.1 [dev] Model is licensed by Black Forest Labs. Inc. under the FLUX.1 [dev] Non-Commercial License. Copyright Black Forest Labs. Inc.
IN NO EVENT SHALL BLACK FOREST LABS, INC. BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH USE OF THIS MODEL.
"The dungeon's floors were spotless. That should have been our first clue."
- Jaster Hollowquill, on his exploration of Undermountain (DnD 5th Edition Monster Manual)
One of the most iconic monsters in the history of roleplaying games is probably the Gelatinous Cube from DnD, and while Flux can make a cube of gel and stick a sword in it, I'd say this Lora should do a better job at making an actual monster. I captioned the training data with natural descriptions of the images as the first attempt with standard settings and the comma-separated captions from the autotagger resulted in an utter failure (thanks for letting me waste 2.5k buzz on that steaming pile of garbage, Civitai, there should be a warning about using standard settings for training a Flux-Lora somewhere! :p), so "gelatinous cube" is kinda the trigger (duh) as that was used in every caption.
What else should be working:
- fantasy-style, anime-style, comic-style
- all sorts of different colors
- "the cube is engulfing/encasing x" with x being armor, bones, swords, skulls, people, skeletons, etc.
- in a medieval dungeon or ruin is the most trained background setting, but other backgrounds will probably work, too. cobblestone floor was pretty much everywhere in the training data and you might even get it when you don't ask for it. If the Lora refuses to comply, lowering the weight should help (see the spaceship example image).
Likes, reviews, comments, hints and tips about training are all very welcome and please, let me see the stuff you create with it! 😁