Lumachrome (Illustrious)

https://redd.it/1sgg02j
@rStableDiffusion

I made an open source alternative to Higgsfield AI

https://redd.it/1sgrgef
@rStableDiffusion

Batch caption your entire image dataset locally (no API, no cost)

I was preparing datasets for LoRA / training and needed a fast way to caption a large number of images locally. Most tools I used were painfully slow either in generation or in editing captions.

So made few utily python scripts to caption images in bulk. It uses locally installed LM Studio in API mode with any vision LLM model i.e. Gemma 4, Qwen 3.5, etc.

GitHub: https://github.com/vizsumit/image-captioner

If you’re doing LoRA training dataset prep, this might save you some time.

https://redd.it/1sgk4yu
@rStableDiffusion

Outside of training a Lora what do people do to keep a face looking correct when making edits to an image?

Mostly been using Klein and Qwen. As per the title, if you change positions, angles of the person in the starting image too much, they lose the likeliness. I've tried using a close up of the face as a 2nd image reference, and tried inpainting on a second pass. Any other ideas?

There's also a Best Face Swap lora which I thought might work but with the same face, but nope.

https://redd.it/1sgvzed
@rStableDiffusion

Light Novel style book illustrations with anima-preview2

https://redd.it/1sgvi4v
@rStableDiffusion

Updates to prompt tool - First-last frame inputs - Video input - Wildcard option, + more

https://redd.it/1sgtopq
@rStableDiffusion