horrible results when using sd 1.5

[loboere]

I'm trying to run the base model without any dreambooth lora but the results are random or meaningless stuff

.yaml config

NewModel:
  path: ""
  base: ""

  motion_module:
    - "models/Motion_Module/mm_sd_v14.ckpt"
    - "models/Motion_Module/mm_sd_v15.ckpt"
    
  steps:          25
  guidance_scale: 7.5

  prompt:
    - "a girl"

  n_prompt:
    - ""

if something is wrong can you tell me how to configure it well for sd 1.5

[sciart17]

I have met the same problem. The result generated by sd1.5 is wired.

[vicentowang]

I'm trying to run the base model without any dreambooth lora but the results are random or meaningless stuff

.yaml config

NewModel:
  path: ""
  base: ""

  motion_module:
    - "models/Motion_Module/mm_sd_v14.ckpt"
    - "models/Motion_Module/mm_sd_v15.ckpt"
    
  steps:          25
  guidance_scale: 7.5

  prompt:
    - "a girl"

  n_prompt:
    - ""

if something is wrong can you tell me how to configure it well for sd 1.5

see the training procedure, motion_module are connected with the base model, must be used

[rlawjdghek]

@vicentowang The weight forms of dreabooth and stable diffusion are exactly the same. So, if the code to load the custom t2I is not executed, unet will be initialized with stable diffusion v1.4(5). The code to load the motion module works completely independently and separately from this process.

[haofanwang]

Can you post result here?

[rlawjdghek]

The result of sd1.4 instead of DREAMBOOTH:

[zengziru]

I have met the same problem.