Nunchaku fixed lightning loras for baked-in Qwen Image Edit 2509 INT4/FP4 distills – visible improvement in prompt adherence with 251115 version
https://redd.it/1p4bdd7
@rStableDiffusion

Raylight Parallelism Benchmark, 5090 vs Dual 2000 Ada (4060 Ti-ish). Also I enable CFG Parallel, so SDXL and SD1.5 can be parallelized.
https://redd.it/1p4je4t
@rStableDiffusion

Wonder Festival Projection Mapping with AI: Behind The Scenes + Workflows
https://redd.it/1p4jqyn
@rStableDiffusion

Are there methods of increasing image generation speed for SDXL models?

I saw this: https://civitai.com/models/1608870/dmd2-speed-lora-sdxl-pony-illustrious?modelVersionId=1820705 and found about Lighting and Hyper models, but I cannot change to another model as none of my loras will woork with it, retraining over 50 loras isn't doable...

But other than Sage Attention, which I just can't get to build I saw that there might be many ways of increasing speed or using less steps for some gens like with video models, what do you guys know out there?

I'm mainly a illustrious user since its better than pony at non-real-life concepts and loras.

https://redd.it/1p4lhhv
@rStableDiffusion

Full Music Video generated with AI - Wan2.1 Infinitetalk
https://www.youtube.com/watch?v=T45wb8henL4

https://redd.it/1p4rd1f
@rStableDiffusion

Control Your Light With The Multi Light LORA for Qwen Edit Plus Nunchaku
https://youtu.be/m5cCzcPwtPA

https://redd.it/1p4u2lu
@rStableDiffusion

High Variation Qwen Workflow

https://redd.it/1p4pv8w
@rStableDiffusion

Chroma HD

https://redd.it/1p4zoms
@rStableDiffusion