Lametta Old

🚀 Old Model Repository

This repository serves as a storage for models that have been previously uploaded. The idea is that once a model is uploaded, it should be accessible to everyone at any time, as long as there are no issues. Here, you can find old models, discontinued models, and miscellaneous merged models for fun, outside of the main models.

✨ Features

Material (material_24821 - 24x14)

• Purpose: This is a merging material. When you merge about 0.5 of this material with other models, the resulting model will have a "lametta - like" effect. The merging ratio depends on the target model, so you can freely adjust it or perform hierarchical merging as you like.
• Output characteristics: Since it's like the core part of lametta, using it directly will result in a strongly deformed output. It's a material model for obtaining outputs with strong deformation and various omissions, so the fineness of the target model may be lost.
• License: The license follows the same rules as lametta, as described at the end of this model card.
• Composition:
  • v1745 + (littleMonsters_anime - coma_v2) = A
  • A + angelbreed_v30 = B
  • B + BracingEvoMix_v2 = C
  • C + (calicomix_v85 - reLIVE_original) = D
  • B + v2012 = E
  • D + (E - weakenedUnity_v60) = material_24821 - 24x14
  • VAE = zVae_v10
  • Note: B corresponds to lametta_v2192test which was uploaded without any explanation, and C is lametta_v21121test. D is adjusted to have a minimum description by receiving (E - weakenedUnity_v60) which is even more matte.

Examples of Merged Works

A large number of examples of merged works are provided below. Some models may produce better results if fine - tuned through hierarchical merging. Note that some of the target models may be difficult to obtain now, and we cannot solve this problem if you contact us.

Future Plans

It's unlikely that new models will be released in the future. It's been almost a year since the previous major release (v2012). Although no new models have been released, it doesn't mean that nothing has been done. A large number of models have been created and discarded through continuous merging (according to the SuperMerger log, about 350,000 merges). Even if an interesting model is created, if a small flaw is found, it will be shelved. Considering that the SDXL environment is already mainstream and more advanced models are emerging, it might be the right time to stop here. I'd like to express my gratitude to those who said "Your models are quite good" in an inconspicuous way. Thank you very much.

Old and Discontinued Models

v21240

• Usage note: To avoid the nan error, explicitly set the VAE when using this model. liquid111vae_v10 or zVae_v10 is recommended.
• Model information: It is the 240th merged model in the v21 series, merged in February 2024. According to the memo, it readjusted the merge distribution of v1745 + (littleMonsters_anime - coma_v2) and remerged it with angelbreed_v30 with the intention of readjusting v2192.
• Image: [01435 - 791397850.png](https://cdn - uploads.huggingface.co/production/uploads/64172e2f1f1f3b0fa80ce889/wdqU7HLA1XFaoDDr3Zfe9.png)

v21121test

• Model information: It is the 121st merged model in the v21 series. According to the memo, it is an improved version of v2192 by hierarchically merging BracingEvoMix_v2. It was more of a step towards creating a better model rather than being released directly.
• Image: [01416 - 547717475.png](https://cdn - uploads.huggingface.co/production/uploads/64172e2f1f1f3b0fa80ce889/5rBkmSL - 9vy5UQ2vPb68Z.png)

####... (Other models are described similarly)

Flat Deformation (FD)

FDv111, FDv109, FDv108

These models were made by incorporating LoRA and other things from 2vXpSwA7 to achieve a flat - deformed output. It seems that around this time, the idea of creating an anime - style model started to emerge.

• Image: [02153 - 453981126.png](https://cdn - uploads.huggingface.co/production/uploads/64172e2f1f1f3b0fa80ce889/rgbKbaJp6sOPXYrsgRdD6.png)

Squinted - Eye Model (J)

v1574J

This is a fun model created by merging v1574 with the half - squinted LoRA of 2vXpSwA7 at a weight of 0.6. Due to its instability in some compositions, it's recommended to use it just for fun. In fact, it's better to use the half - squinted LoRA directly and adjust the prompt, or leave it to a model dedicated to squinted - eye themes.

• Image: [01939 - 681702549.png](https://cdn - uploads.huggingface.co/production/uploads/64172e2f1f1f3b0fa80ce889/xuKK4WIj_0Zvv5QacF0g8.png)

Kemono Model (FU)

FU - v013 (Rejected)

The concept of "kemono degree" varies. Most models can produce a relatively low - level kemono effect, so a dedicated model is expected to produce a high - level kemono. This model was mediocre in this regard and the merging was also immature.

📄 License

All the uploaded models comply with the [creativeml - openrail - m](https://huggingface.co/spaces/CompVis/stable - diffusion - license) license. You can search for "creativeml - openrail - m" to find translated explanations.

As a supplement to Attachment A, using the model to imitate specific works or styles in a way that causes trouble to the right - holders is prohibited.

Usage Permissions