RunDiff+IkiGi FX (M 05)

RunDiffusion FX brings ease, versatility, and beautiful image generation to your doorstep. Join us on our Discord: https://discord.com/invite/wH6dTyBpCf RunDiffusion.com - Powerful Cloud Servers hosting Automatic1111, InvokeAi and other open source tools. The RunDiffusion FX Model Series is a game-changer in the realm of image generation. Balancing simplicity and functionality, this model transforms your brief prompts into diverse, imaginative images. Its versatility caters to a spectrum of creative needs, from the novice enthusiast to the seasoned artist. Above all, it's the ability to produce striking outputs that truly sets it apart. See more image examples here. How to use the model We set out to create a model that can accept small prompts yet generate amazing results. We've done that with RunDiffusion FX. Watch our guide on YouTube. Small prompts work great! (hyperrealism:1.2), BREAK fish swimming under the water BREAK (8K UHD:1.2), (photorealistic:1.2) Use a Negative! e.g. Negative Prompt: plain background, boring, plain, standard, homogenous, uncreative, unattractive, opaque, grayscale, monochrome, distorted details, low details, grains, grainy, foggy, dark, blurry, portrait, oversaturated, low contrast, underexposed, overexposed, low-res, low quality, close-up, macro, surreal, multiple views, multiple angles Wide/Tall aspect ratios are awesome! Resolutions like 480x832 832x480 work great then you can upscale. Lower that CFG! You can use a CFG scale of 3.5 to 5 to get some softer photorealistic images. Olivio Sarikas Likes it! We did not ask him to create this video. This was a surprise to us! License RunDiffusion supports collaboration and innovation, therefore we are releasing this model free and open for all to use. There are only a few simple rules we'd ask you to adhere to: Always use the name of the model in any generation service. "RunDiffusion FX - Photorealistic" "RunDiffusion FX - 2.5D" Do not change the name or obfuscate it. Credit us in any merges Do not resell the model itself -------------------------------------------------------- IkiGai This model is made to achieve the skin detail seen from my initial model Ikigai 2.5D paired with the Danni 2D lora that's no longer available, without any lora needed! V2: Toned back the mix a bit, added in Real Moon. Also integrated a lora I used in Ikigai 2.5D for detail that no longer exists on the site. My usual settings: Sampler: DPM++ 2M Karras Sample Steps: 20-35 Upscaler: ESRGAN_4x Deniose: 0.5 CFG: 5-9 Model has no integrated VAE. I used Grapefruit VAE for the images above. By no means is this law, feel free to experiment and I'd love to see your results!! For inpainting, I generally limit to Only Masked to 800 x 800 or lower and 0.5 denoise (or lower if you want to maintain the initial face look more). It will start to get a little too detailed at higher res'. You shouldn't need to change your prompt at all when inpainting faces to get a good result. This was a goal of mine, as I prefer not to have to do any extra steps in my workflow if possible. This mix comprises of: MajicMix Realistic MIX-Pro-V4 Babes 2.0 FantasticMix Same as the 2.5D model, this one plays very well with 2D artist style loras! Enjoy