Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
A Multipurpose toolkit for managing, editing and creating models.
| Date | Stars |
|---|---|
| 2026-07-31 | 529 |
| 2026-08-04 | 529 |
| 2026-08-06 | 529 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# stable-diffusion-model-toolkit A Multipurpose toolkit for managing, editing and creating models.  Install by: `Extenstions tab > Install from URL > Paste in this pages URL > Install` ## Features - Cleaning/pruning models. - Converting to/from safetensors. - Extracting/replacing model components. - Identifying/debugging model architectures. # Example Many models being distributed are quite bloated, most of their size being redundant or useless data. For example, Anything-v3.0 is 7.7gb and requires a separate 800mb VAE. These can be combined and cleaned into a **2.1gb standalone model**, with the correct VAE included. Easy way, replace the VAE directly: ``` 1. Select Anything-v3.0.ckpt from the source dropdown, press Load. 2. Change to the Advanced tab. 3. Select the class VAE-v1 and leave the component on auto 4. Select Anything-v3.0.vae.ckpt from the import dropdown, press Import. 5. Change the model name to something appropriate like Anything-v3.0.safetensors 6. Press Save. ``` Hard way, build the model from components: ``` 1. Select Anything-v3.0.ckpt from the dropdown, press Load. 2. Change to the Advanced tab. 3. Select the class CLIP-v1, press Export. 4. Select the class UNET-v1, press Export. 5. Press Clear 6. Select NEW SDv1 from the source dropdown, press Load. 7. Select the class CLIP-v1, and select the Anything-v3 CLIP (just exported), press Import. 8. Select the class UNET-v1, and select the Anything-v3 UNET (just exported), press Import. 9. Select the class VAE-v1, and select the Anything-v3 VAE, press Import. 10. Change the model name to something appropriate like Anything-v3.0.safetensors 11. Press Save. ``` ## Advanced The advanced tab lets you replace and extract model components, it also shows the detailed report. Import can extract components from full models, so if you want to replace the CLIP in your model with the SD 1.4 CLIP then you can simply specify the CLIP component and import the SD 1.4 checkpoint. The report will show all matched architectures, all rejected architectures (and reasons why they were rejected), and the list of all unknown keys. This is mostly useful for debuging models to see why they wont load. ## Autopruning Toggle the `Enable Autopruning` option in settings. On startup of the WebUI everything in the `models/Autoprune` folder will be pruned into FP16 `.safetensor` models (except VAEs which will be `.pt`) and moved into their proper folders. Broken or unknown models will be skipped. Models will be renamed if there is a conflict when moving (`NAI.safetensors` to `NAI(1).safetensors`, etc). The `Reload UI` button will also trigger the Autopruning. ## Metric During analysis a metric is computed which attempts to uniquely identify a models weights. The AnythingV3 model produced above has the metric: `(2020/2982/0130)`, which corresponds to `(UNET/VAE/CLIP)`. This toolkit knows the metrics for a few common components and will include any matches in its report. So for `(2020/2982/0130)` it knows the VAE metric of `2982` corresponds to the NAI VAE and will report `Uses the NAI VAE`. Actually many VAEs being distributed are just NAI VAE renamed, without a metric it would be difficult to know. Though this system isnt foolproof and incorrect matches can happen. ## Notes Some things that may be useful to know when manipulating models. ### Components Stable Diffusion requires 3 different components to function: the VAE, the UNET and CLIP. Checkpoints contain all these components. The toolkit can still recognize checkpoints that are missing components, but they wont be considered intact. For example a checkpoint thats missing its CLIP will be recognized as containing the `UNET-v1-BROKEN` and the `VAE-v1-BROKEN` models, which can be exported like normal if you want to fix the model. Any checkpoint under 2gb will be missing a component (or multiple). **T
Excerpt of 6,618 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:7ce44c42a820a2a8, name:stable diffusion
