The pony version of a semi-realistic merge designed for generating good results from NSFW concept LoRAs and transformations. I personally use this model to test out my pony LoRAs. I've been sticking with this one for a while so I felt it's time to post it so that the reference images for some of my LoRAs have the model linked and available. <h3 id="generating-9ryy0g375">Generating</h3>I found the best results were starting with 864x1152, clip skip 2, cfg 7-8, Euler A. Then img2img: resize by 2x, denoise 0.15, Script: SD Upscale -&gt; R-ESRGAN 4x+ @ scale factor 2. (I use forge to help with vram).<h3 id="prompts-562clrd8y">Prompts</h3>Suggested general prompt:score_9, score_8_up, score_7_up, realistic,Suggested img2img prompt additions:ultra realistic, (high quality, detailed, beautiful), detailed beautiful eyes, realistic, outstanding,&lt;lora:add-detail-xl:0.7&gt; &lt;lora:xl_more_art-full_v1:0.1&gt;, &lt;lora:Realistic000003_1_2.5DAnime_0.05_Merge1:0.5&gt;<h3 id="tips-jt1v2nfoi">Tips</h3>This is a pony model so it will mostly generate nudity unless you specify clothing or things like "covered nipples".Using "anthro", "furry", or "scalie" will make the results less human.Specifying a hair color, facial features, or expressions (like "ahegao") will make the results more human.All of the example images were generated without the "ultra realistic" tag onwards. The following tags were added for img2img. I'd suggest leaving realism and detail LoRAs out of the initial generation since they usually dampen concepts.It has a slight watermark bias, you can add "watermark" and "signature" to the negative prompt.

Files with SHA256: 253d41737d96f6df3fc3928d702d2c77d05c53fa295993bb8ba7f7f739f402f1