patrickvonplaten's picture
license: creativeml-openrail-m
base_model: runwayml/stable-diffusion-v1-5
- stable-diffusion
- stable-diffusion-diffusers
- image-to-image
- diffusers
- controlnet
- jax-diffusers-event
inference: true
# controlnet- tsungtao/controlnet-mlsd-202305011046
These are controlnet weights trained on runwayml/stable-diffusion-v1-5 with new type of conditioning. You can find some example images in the following.
prompt: a living room with a dining table and chairs
prompt: a living room with couches and a tv
training datasets:
SD model base on: runwayml/stable-diffusion-v1-5
for the training data, you can find it has 2 colums, raw training data, conditioning training data and its prompts.
raw training data is crawling from the internet for the living room with special training purpose. and they were resized with the local tool, it is the standerlization
process before training.
conditioning data was creating base on raw training data, we deployed the local Stable diffustion and Controllnet plugin on our local Dev Env for this purpose, to setup
the target conditioning data. after this, the conditioning data also be resized with the standardliaztion procedure.
for the prompts, we add this by the manual way, since the small testing data sets. anyway it also can be done by some favour tool on the internet.
above all, we have a upload scripte to combine all the raw data/conditioning data/prompts into a dataset and then auto upload they to Huggingface.
from the conditioning data, you can see which is the framework of the raw data. so training base on this, it will get a mode to extract the line frame work of the input.
and at the same time, raw data are all with some special style. so base on this model, you will get your raw input living room data line framework and then change the input
living room to a special style living room.
from the functional point of view, it seems like "MLSD" model, but it works better on special style living room data.