Hello everyone,
This Checkpoint Merge aims for a photorealistic style. It works well with abstract topics aswell.
Checkpoints I used! ⬇️
CyberRealistic from Cyberdelia
epicRealism - Natural Sin RC1 from epinikion
Photon from Photographer
Special thanks for this amazing Checkpoints! 🙌
I have used these settings almost exclusively. They are a good starting point and can be adjusted depending on what you want to generate.
Sampling method:
DPM++ 2M Karras
DPM++ 2M SDE Karras
DPM++ 2M SDE (if you want even more skin texture, but be careful with it > higher sampling steps can break it and give you a bad result)
Sampling Steps:
20-40
CFG Scale: 3-7 (5 is a good starting point)
I recommend clip skip 1.
So far, I haven't really tried Clip Skip 2. Feel free to try it out and see if it produces good results.
Recommended Upscaler Settings:
Hires. fix Upscaler:
Hires steps: 10-20
Denoising strength: ~ 0.2-0.5
If you can use Hires.fix for better Results!
I don't use restore faces. If you got a bad face or want to change it use ADetailer for it!
I would greatly appreciate any feedback!
If you like this model, I would greatly appreciate a positive review along with constructive criticism, suggestions for improvement, or anything similar!
Description
Hello dear people! I have been working on v4.0 for some time. I tested 6-7 different versions until I was satisfied. With this update, I have tried to further improve the artistic and abstract components. Additionally, I have tried to make the composition cleaner. Finally, I have also started using multiple negative prompts and adjusting them when I am not satisfied with something in the output. This has definitely had a very positive impact.
I hope you have fun with it!
FAQ
Comments (9)
I am unable to get v3 or v4 to work. I am always met with an attribute error (see below). Just curious if there was anything different you did with these versions that might result in this? I am using invoke 3.5.1
[2024-01-04 22:28:04,932]::[InvokeAI]::ERROR --> Traceback (most recent call last):
File "/workspace/invokeai/venv/lib/python3.10/site-packages/invokeai/app/services/invocation_processor/invocation_processor_default.py", line 104, in __process
outputs = invocation.invoke_internal(
File "/workspace/invokeai/venv/lib/python3.10/site-packages/invokeai/app/invocations/baseinvocation.py", line 669, in invoke_internal
output = self.invoke(context)
File "/workspace/invokeai/venv/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "/workspace/invokeai/venv/lib/python3.10/site-packages/invokeai/app/invocations/compel.py", line 69, in invoke
tokenizer_info = context.services.model_manager.get_model(
File "/workspace/invokeai/venv/lib/python3.10/site-packages/invokeai/app/services/model_manager/model_manager_default.py", line 112, in get_model
model_info = self.mgr.get_model(
File "/workspace/invokeai/venv/lib/python3.10/site-packages/invokeai/backend/model_management/model_manager.py", line 490, in get_model
model_path = model_class.convert_if_required(
File "/workspace/invokeai/venv/lib/python3.10/site-packages/invokeai/backend/model_management/models/stable_diffusion.py", line 127, in convert_if_required
return convertckpt_and_cache(
File "/workspace/invokeai/venv/lib/python3.10/site-packages/invokeai/backend/model_management/models/stable_diffusion.py", line 289, in convertckpt_and_cache
convert_ckpt_to_diffusers(
File "/workspace/invokeai/venv/lib/python3.10/site-packages/invokeai/backend/model_management/convert_ckpt_to_diffusers.py", line 1728, in convert_ckpt_to_diffusers
pipe = download_from_original_stable_diffusion_ckpt(checkpoint_path, **kwargs)
File "/workspace/invokeai/venv/lib/python3.10/site-packages/invokeai/backend/model_management/convert_ckpt_to_diffusers.py", line 1432, in download_from_original_stable_diffusion_ckpt
set_module_tensor_to_device(vae, param_name, "cpu", value=param)
File "/workspace/invokeai/venv/lib/python3.10/site-packages/accelerate/utils/modeling.py", line 269, in set_module_tensor_to_device
new_module = getattr(module, split)
File "/workspace/invokeai/venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1695, in getattr
raise AttributeError(f"'{type(self).__name__}' object has no attribute '{name}'")
AttributeError: 'Attention' object has no attribute 'to_to_k'
[2024-01-04 22:28:04,937]::[InvokeAI]::ERROR --> Error while invoking:
'Attention' object has no attribute 'to_to_k'
Both full version and pruned version are not working? Unfortunately I don't know what the error could be. I don't use InvokeAI. But someone in the comments posted this message. Maybe that helps?
Things based on diffusers don't seem to like it (v2 is fine). I needed to use the following python script to rewrite the .safetensors file before it would work with what I usually use (though its fine in ComfyUI):
(No idea whether civitai will format this correctly (apparently not))
```python
import sys
import os
from safetensors import safe_open
from safetensors.torch import save_file, load_file
def fix_diffusers_model_conversion(load_path: str, save_path: str):
# load original
tensors = {}
with safe_open(load_path, framework="pt") as f:
for key in f.keys():
tensors[key] = f.get_tensor(key)
# migrate
new_tensors = {}
for k, v in tensors.items():
new_key = k
# only fix the vae
if "first_stage_model." in k:
# migrate q, k, v keys
new_key = new_key.replace(".to_q.weight", ".q.weight")
new_key = new_key.replace(".to_q.bias", ".q.bias")
new_key = new_key.replace(".to_k.weight", ".k.weight")
new_key = new_key.replace(".to_k.bias", ".k.bias")
new_key = new_key.replace(".to_v.weight", ".v.weight")
new_key = new_key.replace(".to_v.bias", ".v.bias")
new_tensors[new_key] = v
# save
save_file(new_tensors, save_path)
if name == "__main__":
fix_diffusers_model_conversion(
os.path.normpath(sys.argv[1]),
os.path.normpath(sys.argv[1] + "_fixed"),
)
```
@Fusch 'Attention' object has no attribute 'to_to_k' was the error I was getting, so it looks like the same thing I needed that script for.
@AwayWithTheFaeries That script worked like a charm. Fully functional now. Thank you for sharing
3.0 和奇幻题材结合时想象力爆棚，真是太佩服作者了！
Hey, what's the difference between the versions? is v4 the best?
Very versatile and stable, good work Fusch.
Lo puedo usar en Comfy UI?
Quite a good photorealistic model with good flexibility and versatility. Only sometimes has problems with hand distortion. Anyway, it's a good merge. Good job!
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.