Pony experimental model used for photorealistic generation.

V3: Use DPM++ SDE, CFG of 5 recommended, and 20 to 30 Steps.

Samples are upscaled using Ultimate SD Upscale, with the same prompts and a denoise of 0.1 to 0.25, higher might create artifacts or image repeats. FaceDetailer from Impact pack using Eyeful-V2 or one of the face yolov models.

If it is a close-up portrait, face or eye detailer is likely unnecessary, too much denoise on an upscale may also create issues in close-ups as the model is a little aggressive. Inpainting manually may yield better results.

Click the samples to get the workflows for ComfyUI.

The model tends to make them muscular and in their 20-30 year old range. Add the word "mature" and a weight to age them up. For other body types, add "muscular" to the negatives and the body type to the positives.

V2: Use DPM++ SDE, DPM++ 2M SDE, and Restart. Euler gives the image a softer look. CFG of 7 recommended, but it works well in a large range of CFGs, and 20 to 30 Steps.

V1: Use DPM++ SDE to approach more photo-like results. CFG of 7 recommended, and 20 Steps.

The model responds to Pony Tags, but you can also use it without and prompt like with SDXL.

Using "score_9, score_8_up, score_7_up, score_6_up" in the positives, and "score_4, score_3, score_2, score_1" in the negatives is not necessary, but will yield better results.

Do not overdo the negatives with word vomit, it will most likely decrease the quality of the resulting image. Only use negatives when the result does not match the prompt, like some clothing will trigger a female automatically, use "1girl, woman, female, femenine" in the negatives in this case.