Terms
Stable Diffusion Term Meaning CFG Classifier Free Guidance VAE Variational AutoEncoder
Prompt techniques
Tested in Automatic1111 (and most in ComfyUI). Reference: https://www.youtube.com/watch?v=MgLpmksylOc&t=482s Syntax Example Behaviour [a|b] [cat|dog] Alternatively will use cat and dog for every step [a|a|a|b] [cat|cat|cat|dog] Will use cat for every step, except each fourth step, which will be dog [from:to:when] [cat:dog:10][cat:dog:.6] Will use cat for the first 10 steps and then switch to dogWill use
Stable Diffusion samplers
Name Effective max steps Remarks Euler 30 Euler adaptive 100 Heun 30 LMS 50 PLMS 60 DDIM 60 DPM adaptive 10 I don’t use it DPM fast 150 DPM 2 40 DPM 2 adaptive 100 DPM 2 adaptive Karras 100 DPM 2 Karras 30 I don’t use it
Alternatives
Chat: Prompters: Bots:
Lora training cheat sheet
References https://www.youtube.com/watch?v=N_zhQSx2Q3chttps://www.youtube.com/watch?v=1BCYdd9r1To Tools Install kohya_ss Prepare dataset Image captioning from koyha_ss Configure training in koyha_ss Selecting the best model Cloud generation All of the above is intended to be installed and executed locally, but doing LoRA training in the cloud is a very good – if not better – alternative. I use this provider +
Stable Diffusion: Textual Inversion embeddings
References I have written this article as a personal summary of following detailed videos on the subject: Settings In stable-diffusion-webui go to: Settings -> Training and preferably enable following (RAM allowing): If your video card does not have enough VRAM, you could try and disable following setting (note that this will decrease training success rate):
Stable Diffusion: VAE
VAE stands for Variational AutoEncoder. Images are generated in a “compressed” way, with may result in artifacts. Models may come with a integrated VAE, but can also benefit from a custom one. You can install VAE files under models/VAE by downloading the applicable file there. Two VAE’s can be found on https://huggingface.co/stabilityai/sd-vae-ft-mse-original/tree/main, of which I
Stable Diffusion: Installation with web UI
Base installation on host Make sure you have Git and a compatible Python version (currently 3.10) installed and available in your path. Open a command prompt and execute the command below in a directory of your choice. Note that this will create a subdirectory called stable-diffusion-webui. git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git Download a model Stable diffusion needs
Find git origin URL
Since I always seem to forget the command…
Stable Diffusion: Model hashes
Overview of the model hashes I use in Automatic1111 Name Type Trigger Hash (SHA256) Anything-V3.0-pruned ckpt 543BCBC21294831C6245CD74C8A7707761E28812C690F946CB81FEF930D54B5E Anything-V3.0-pruned safetensors 6806D4C0F86A2F39D60A97BCF926F39D8B8FCE2C71E39BAF4EF0EE40A5655632 anything-v4.0-pruned safetensors 69528490DF7C34087AE45CD2B36BC8CA6CE88C51A36392F47608A51C6FF4A7C7 anything-v4.5-pruned safetensors 6E430EB51421CE5BF18F04E2DBE90B2CAD437311948BE4EF8C33658A73C86B2A f222 ckpt 9E2C6CEFF3F6D6F65C6FB0E10D8E69D772871813BE647FD2EA5D06E00DB33C1F mdjrny-v4 ckpt 5D5AD06CC24170B32F25F0180A357E315848000C5F400FFDA350E59142FABD68 sd-v1-5-inpainting ckpt C6BBC15E3224E6973459BA78DE4998B80B50112B0AE5B5C67113D56B4E366B19 unstableinkdream_v6 safetensor nvinkpunk,kuvshinov,dreamlikeart,samdoesart,analog style,modelshoot style D855C1D2BABB485F5C1DE91D2EA28F295E72DDA052A7C8C04ADFF16E3013988B v1-5-pruned-emaonly ckpt CC6CB27103417325FF94F52B7A5D2DDE45A7515B25C255D8E396C90014281516 v2-1_512-ema-pruned ckpt 88ECB782561455673C4B78D05093494B9C539FC6BFC08F3A9A4A0DD7B0B10F36 v2-1_768-ema-pruned ckpt AD2A33C361C1F593C4A1FB32EA81AFCE2B5BB7D1983C6B94793A26A3B54B08A0 Hashes calculated with Powershell “Get-ChildItem