Verified: 6 months 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.
Strength 0.5-1.2, varies with style, prompt complexity, sampler, and how many other LoRAs you are using. Rough rule- higher strength is more Hollywood/Celebrity Squares and lower is more flexible (semi-unstable comedy IMO). Start at .8 or .9.
Trigger tags- italics are variables for you to choose, straight brackets are optional
a square 3x3 grid of 9 square pods with value in each square, [with names at the bottom]
a square 3x3 grid of 9 squares with {value} in each square
a contestant area with three podiums, [the host at a raised center podium]
a square pod portrait depicting value, [sitting in a pod labelled as "value"]
host
Notes
Names are usually nonsense words, but you can list the names and it works (mostly)
The larger resolution your system can handle, the better. Does well with 1-2 mpx. Copy this Google Sheet and you can edit to find ratios: https://docs.google.com/spreadsheets/d/1p913YOU9A6rC0nasQPvKWsNDrE-OOUHU4-AZI8Eqois/edit?gid=0#gid=0
this is super flexible and you do not need to stick with the suggested triggers. Show me what you create!
squares vs. square pods: With people, "square" tends to be a more comedic result than "square pod". If your human subjects do not have legs, "square pod" helps, as does "sitting" or "seated".
Grid has the wrong number of squares- try switching between "squares" and "square pods" ot sticking a comma right after "pods" or "squares". Roughly 1 out of 5 generations have this issue, even when following the format I specific. I retrained a few times to address. Having multiple subjects really throws Flux off. Too strong of training, it results in it thinking 9 isn't enough, less training and adjusted LR/noise offsets got the square count consistent but text and general quality was meh
Should be fine to use onsite- all images used in the training were created with prior versions of the LoRA. Celebrity names may appear but I went out of my way to remove any likeness.