<a target="_blank" rel="ugc" href="https://rundiffusion.firstpromoter.com/login">Trained with OSTRIS AI-Toolkit @ </a><a target="_blank" rel="ugc" href="http://runpod.io">runpod.io</a> Flux3 Klein: Flux: Best used with the base model — 50 steps, CFG 3.5–4. Wan: WAN: I haven’t created any videos with the LoRA, so I can’t judge how well it works as a video LoRA. But in the WAN refiner, it’s fantastic for upscaling images and adding more details. Qwen: My impressions: The paint looks super good, just like the backgrounds, and the fonts are great. The rims also look consistently better compared to Flux, except for the logo on the rim. However, Qwen struggles a lot with fine details like sensors and logos. HiDream: This is my first HiDream LoRA, trained with the default template settings. My first impression: it’s not as great as Flux yet, but I’m curious to see how the model develops. I’ll keep working on optimizing the training. The fully open-source license is especially interesting for commercial use.

Porsche Targa [Flux 3,Qwen, WAN, Hidream, Ideogram]

Flux3_Targa.safetensors

Files with SHA256: 2a641f61acae736f8d9605e2fdc6f4f2c4b14b39ed477218b30fc742b4397de0