6.5 is the latest. You can think of 6.5-clip-fix as an optional patched version of 6.5 that may or may not improve the models output. The term "fix" is misleading, as it implies that the non-fixed version is broken, it is not. 6.5 has a max deviation of: 0.00001compared to 6.5-CLIP-Fixed: 0.00000. This slight deviation will not result in any corruption of the generated outputs. I included the clip-fix version as it does slightly change output. I personally find the "fixed" version is less creative and looks too much like standard SD v1.5. That's why I put it in the middle, as an option for those who want it. I'll update the description to better explain the difference between the two versions as it is a bit confusing.

@rockedt No problem! Glad I could help. And thanks for the feedback! I've been working with machine learning and neural networks for 10+ years so I often forget that others understandably, aren't aware of all the technical jargon lol.

Do you have Hires fix and restore faces enabled?
I believe I used:
Restore Faces: True(on)//CodeFormer//0.8
Hires.Fix: Ture(on)//None(upscaling)//0.7(denoise)

I also have my ETA noise seed delta set to: 31337 (You can find this in the webui settings menu)

Not exactly. The CLIP issue stems from merging via the 'Checkpoint Merger` in AUTOMATIC1111.
Try this:
1. Make sure your dreambooth models are trained on SD v1.5 and not 2.0+
2. Install this extension to webui - https://github.com/iiiytn1k/sd-webui-check-tensors
3. Load your dreambooth model into check-tensors extension
4. It should return an int64 tensor matrix
* Good clip should look like:
tensor([[ 0.0000, 1.0000, 2.0000, ...., 76.0000]])
5. Broken clip would deviate, for example:
tensor([[ 0.0452,1.6535, ..., 75.9852]])
6. Dreambooth models shouldn't have any deviation.
7. Install the following extension in webui - https://github.com/bbc-mc/sdweb-merge-block-weighted-gui
8. Merge your fine-tuned model(A) with the bigger model(B)
9. Name your output model
10. At the bottom of the merge-block-weights extension, you'll see a slider named "M00" Which stands for "Middle Block".
11. Set M00s value ( I recommend leaving it at it's default position for initial testing) and don't worry about "IN00", "OUT11", ect.. These are UNet blocks and can be left at their default values for your purpose.
12. IMPORTANT In the Merge-block-weight extension, to the right, you'll see a box called "Skip/Reset CLIP position_ids" Select Force Reset and then Run Merge
13. Switch over to the check-tensors tab in webui
14. Load your newly merged model and inspect the returned tensor matrix for deviation, as outlined earlier.

Extra: You can also use this extension - https://github.com/arenatemp/stable-diffusion-webui-model-toolkit to check for broken CLIP.

Hope this helps, good luck!

If you're using AUTOMATIC1111 webui, then yes you can.
You'll need an inpainting model and/or a pix2pix model. I recently added both model types for this model, but really you can use whichever you prefer.
1. After you generate your image with Txt2Img, under the image preview you will see a few buttons that say Send To: <inpainting> <Img2Img> <Extra>
2. If you are using an inpainting model, choose inpainting. If you're using Pix2Pix, choose Img2Img
3. Select and load the inpainting or Pix2Pix model.
4. If inpainting, use the brush to mask to the part of the image you want to change. Then write a prompt describing the thing you want to change.
5. If Pix2Pix, you can simply tell the model what you want it to do inside the prompt field. For example, if the subject is wearing a blue shirt but you want it to be red, type something like; "Make her shirt red".

Hope this helped!

@Skibidybap No problem, glad to help!