Stable Diffusion Problems? Inpainting, Model Loading, Python, Forge and Error 128

Stable Diffusion can be frustrating when an inpainted area looks worse than the original, a model refuses to load or the WebUI suddenly stops launching after an update. Most of these problems are fixable, but the solution depends on the exact message shown in the terminal.

Before changing or deleting anything, copy the full error and read the few lines above it. The final line often says only that the process failed, while the real cause appears earlier.

1. Inpainting Denoise Washout Too Low

The Stable Diffusion inpainting denoise washout too low issue usually appears when the masked area looks faded, blurry or almost unchanged. In most cases, the denoising strength is too low for the amount of editing you are asking the model to perform.

A low setting tells Stable Diffusion to preserve the original image. That is useful for removing a small blemish, but it may not be enough when replacing clothing, changing facial details or removing an object.

How to Fix It?

Increase the denoising strength gradually rather than jumping straight to 1. Start around:

• 0.25–0.40 for small repairs
• 0.40–0.60 for faces, hands or clothing
• 0.60–0.80 for replacing an object completely

If the result still looks washed out, reduce the mask blur slightly. A wide mask blur can soften the edited area so much that it loses contrast.

For small details, choose Inpaint area: Only masked. This gives Stable Diffusion more resolution to work with inside the selected area.

2. Model Failed to Load

The Stable Diffusion model failed to load error usually means the WebUI cannot find, read or initialize the selected checkpoint.

This often happens after downloading an incomplete model, placing it in the wrong directory or trying to load a model that is not compatible with the current setup.

How to Fix It?

First, confirm that the checkpoint file ends in .safetensors or .ckpt.

For AUTOMATIC1111, the usual checkpoint folder is:

stable-diffusion-webui\models\Stable-diffusion

Place the model there, restart the WebUI and refresh the checkpoint list.

If it still fails, try loading a different checkpoint. When another model loads correctly, the original file is probably incomplete, corrupted or incompatible.

Also check the terminal for messages such as:

• FileNotFoundError
• CUDA out of memory
• size mismatch
• UnpicklingError
• No checkpoints found

A memory error requires different action from a missing-file error, so do not reinstall everything before reading the traceback.

4. Inpainting Fill Mode Creates Awkward Frames

The Stable Diffusion inpainting fill mode creates awkward frames problem usually appears as a visible rectangle, strange color block or unnatural border around the edited area.

Fill mode uses nearby colors to create the starting content inside the mask. This can work well for simple backgrounds, but it may struggle with complicated lighting, textures, faces or perspective.

How to Fix It?

Change Masked content from Fill to Original when repairing an existing subject.

Use Latent noise when adding a completely new object.

You should also:

• Expand the mask slightly beyond the unwanted object.
• Increase the masked-area padding.
• Use Only masked for small edits.
• Reduce mask blur when the edges look too soft.
• Increase denoising strength if the original object remains visible.
• Lower denoising strength if the replacement looks disconnected.

If the awkward frame follows the exact shape of the processing area, the problem is often caused by insufficient padding or an unsuitable crop size.

5. WebUI Forge

Stable Diffusion WebUI Forge is based on the familiar Stable Diffusion WebUI layout but focuses heavily on performance, memory management and newer generation tools.

Forge is usually easier to install through its official Windows package because the required Python and Git components are included.

How to Install or Repair Forge?

Extract Forge into a simple folder such as:

C:\Forge

Avoid installing it inside Program Files, OneDrive or a deeply nested folder.

If Forge stops working:

1. Close the launcher.
2. Disable recently added extensions.
3. Run the update file included with Forge.
4. Launch it again without custom arguments.
5. Check whether the checkpoint path is correct.

Do not copy the entire venv folder from AUTOMATIC1111 into Forge. Even though the interfaces look similar, their environments and dependencies may differ.

You can share model folders between installations, but it is better to configure the model path rather than duplicating or moving the entire installation.

6. Inpainting Denoise Strength

Stable Diffusion inpainting denoise strength controls how much freedom the model has to change the masked area.

A low value keeps the result close to the original. A high value lets the model rebuild the area more aggressively.

There is no single perfect setting, but these ranges are useful starting points:

• 0.20–0.35 for blemishes and tiny corrections
• 0.30–0.50 for eyes and facial details
• 0.40–0.60 for hands, fingers and clothing
• 0.55–0.75 for replacing an object
• 0.70–1.00 for generating new content inside an empty area

Higher denoising does not automatically mean higher quality. It only means more change.

When the result looks almost identical to the original, raise the value. When the face, pose or lighting changes too much, lower it.

7. Error 128

The Stable Diffusion error 128 message usually appears when Git fails to download or update a required repository.

The same applies to error code 128 Stable Diffusion. The number itself is not very useful because Git uses it for several fatal errors. The important information is the message immediately above the exit code.

Common causes include:

• Git is missing.
• Git is not available in PATH.
• GitHub is blocked by a VPN, proxy or firewall.
• The internet connection was interrupted.
• A previous clone is incomplete.
• The installation folder is protected.
• Antivirus software blocked a file.
• The disk is full.

How to Fix Error 128

Open Command Prompt and run:

git --version

If Windows does not recognize the command, install Git and restart the computer.

If Git is installed, check whether an old proxy has been configured:

git config --global --get http.proxy
git config --global --get https.proxy

Remove the proxy only when it is outdated or incorrect:

git config --global --unset http.proxy
git config --global --unset https.proxy

Close Stable Diffusion and delete only the incomplete repository mentioned in the error. Then run the launcher again.

Do not delete your checkpoints, LoRAs, embeddings or output folders.

8. Couldn’t Launch Python

The Stable Diffusion couldn't launch Python error means the launcher cannot find or start the required Python installation.

This commonly happens when Python was installed without adding it to PATH, Windows is redirecting the command to the Microsoft Store or the WebUI is pointing to a Python version that no longer exists.

How to Fix It?

Check whether Python works:

python --version

You can also view all detected installations with:

py -0p

For many AUTOMATIC1111 installations, Python 3.10.6 remains the safest choice.

When installing Python, make sure you select:

Add Python to PATH

If typing python opens the Microsoft Store, open Windows App execution aliases and disable the aliases for python.exe and python3.exe.

You can also point the WebUI directly to Python by editing webui-user.bat:

set PYTHON=C:\Path\To\Python310\python.exe

If the existing virtual environment was created with the wrong Python version, close the WebUI, delete the venv folder and launch it again. The launcher should rebuild the environment.

9. Fix Long Necks in Image

To Stable Diffusion fix long necks in image, use inpainting rather than regenerating the complete picture.

A long neck usually appears because the model loses the correct relationship between the head, shoulders and upper body. Masking only the neck may not provide enough context to repair it.

How to Fix It?

Mask the full neck, part of the shoulders and a small section of the upper chest. Keep the face outside the mask when you want to preserve the character’s identity.

Use:

• Inpaint area: Only masked
• Masked content: Original
• Denoising strength around 0.40–0.55
• Enough padding to include the chin and shoulders

Prompt example:

natural neck length, proportional shoulders, realistic collarbone,
balanced upper body, anatomically correct posture

Negative prompt example:

long neck, stretched neck, disconnected shoulders, deformed anatomy,
unnatural proportions, distorted upper body

If the first attempt changes the face, reduce denoising strength or tighten the mask around the lower chin, neck and shoulders.

Sometimes the best approach is to repair the neck and shoulders first, then correct the face in a second pass.

10. RuntimeError: Couldn’t Clone Stable Diffusion

The RuntimeError: couldn't clone Stable Diffusion message appears when the launcher tries to download a required repository but Git cannot complete the operation.

This is often connected to Error 128, but the terminal should contain a more specific message explaining what went wrong.

First, check Git:

git --version

Then test whether cloning works manually:

git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git

The result usually points to the cause:

• Could not resolve host means a DNS or connection problem.
• Connection timed out may indicate a VPN or firewall issue.
• Permission denied means the folder is not writable.
• Destination path already exists means an incomplete folder is already there.
• Repository not found means the URL is incorrect or unavailable.
• SSL certificate problem may be caused by system certificates or network inspection software.

Once the real cause is fixed, remove the partial clone and run the launcher again.

For Forge, the official packaged installer is usually easier than manually cloning all required repositories.

Quick Troubleshooting Checklist

When Stable Diffusion stops working, check the problem in this order:

1. Read the complete terminal message.
2. Confirm that Python and Git are working.
3. Make sure at least one valid checkpoint is installed.
4. Disable recently added extensions.
5. Check the VPN, firewall and internet connection.
6. Use a short installation path with normal English characters.
7. Remove only the broken virtual environment or incomplete repository.
8. Reinstall the entire WebUI only when the targeted fixes fail.

For inpainting problems, change one setting at a time. Test denoising strength first, followed by masked content, mask blur, padding and the selected inpaint area. Changing everything together makes it difficult to tell which setting actually solved the problem.