I am using AgnAI with KobolCPP and I am looking for an uncensored Image AI to use with AgnAI. Does anyone have any recommendations? Much appreciated.
- Jul 6, 2017
- 104
- 134
I'm confused on what you're asking for. KoboldCPP/AgnAI are LLM-based AI softwares, but you seem to be asking for software for image generation? The easiest to jump right into is Automatic1111 (Automatic1111 AMD .. if you're on an AMD gpu). The default model that it uses is pretty lame (SD 1.5). I recommend going to civitai and searching for a model that fits your needs, and using it. These are the models I use the most, depending on what I'm going for:
- Anime style: revAnimated (SD) / ponyDiffusion (SDXL)
- Cartoon: samaritan3dCartoon (SD) / (don't have an SDXL cartoon model)
- Realism: dreamshaper_8 (SD) /pyrosNSFWSDXL (SDXL)
- General: chilloutmix (or any chillout model, any of the models listed above also work decently here)
Others that I've used and kept (which means they did acceptable for my liking): bluePencilXL, chilloutrealistic, picrealism_naturalSin, corneos7thHeavenMix, abyssorange, pixreal
About SD vs SDXL: SDXL does far better with fewer iterations (I use 30 for SDXL, and 35 for SD), but it does require a bit more VRAM (8GB min) and time per iteration (a few seconds per iteration).
Regarding censorship: it doesn't matter if you're talking LLM or image generation, censorship is going to be based on the model, not the software that uses the model. For LLMs you would search for an "uncensored" model, and for image generation they usually put "NSFW" in the title.
- Anime style: revAnimated (SD) / ponyDiffusion (SDXL)
- Cartoon: samaritan3dCartoon (SD) / (don't have an SDXL cartoon model)
- Realism: dreamshaper_8 (SD) /pyrosNSFWSDXL (SDXL)
- General: chilloutmix (or any chillout model, any of the models listed above also work decently here)
Others that I've used and kept (which means they did acceptable for my liking): bluePencilXL, chilloutrealistic, picrealism_naturalSin, corneos7thHeavenMix, abyssorange, pixreal
About SD vs SDXL: SDXL does far better with fewer iterations (I use 30 for SDXL, and 35 for SD), but it does require a bit more VRAM (8GB min) and time per iteration (a few seconds per iteration).
Regarding censorship: it doesn't matter if you're talking LLM or image generation, censorship is going to be based on the model, not the software that uses the model. For LLMs you would search for an "uncensored" model, and for image generation they usually put "NSFW" in the title.