This lora is a 'port' for Wan video 2.2 of my previous versions for <a target="_blank" rel="ugc" href="https://civarchive.com/models/495362/alfonso-azpiri-style-for-pony-xl">Pony XL</a> and <a target="_blank" rel="ugc" href="https://civarchive.com/models/22982/a-azpiri-style">SD 1.5</a> of Alfonso Azpiri style, a a mythical Spanish artist well known for his erotic Lorna comics and for creating more than 200 Spanish video game covers in the 80s and 90s. He also published his comics in the prestigious magazine 'Heavy Metal', and also made comics for young audiences like those of his character Mot. Its graphic style is very characteristic and attractive. For this one, I recomend to use 0.7-0.8 strenght for the High Noise lora and 1 for the Low Noise lora. After trying many of the recommended methods with accelerator loras, the one that works best with this model and I recommend using in general with Wan 2.2, is the following: - Use a normal workflow with 2 ksamplers (high noise/low noise phases). - Install the <a target="_blank" rel="ugc" href="https://github.com/ClownsharkBatwing/RES4LYF">RES4LYF custom nodes</a> from the ComfyUI Manager in ComfyUI to be able to use the custom optional samplers/schedulers it includes (res_2s, res_3s, etc... /Beta57, Bong Tangent, etc...)- In the High Noise model phase, use <a target="_blank" rel="ugc" href="https://civarchive.com/models/1585622?modelVersionId=2080907">Lightning 1.1 t2v High</a> 4 step Wan 2.2 lora with 0.75 strenght AND the <a target="_blank" rel="ugc" href="https://civarchive.com/models/1585622?modelVersionId=2014522">lightx2v wan 2.1. i2v Rank64 Fixed</a> accelerator lora (yes, the image2video one) with 2.20 strenght.- In the Low Noise model phase, use <a target="_blank" rel="ugc" href="https://civarchive.com/models/1585622?modelVersionId=2081616">Lightning 1.1 t2v Low</a> 4 step Wan 2.2 lora with 1 strenght AND, again, the same <a target="_blank" rel="ugc" href="https://civarchive.com/models/1585622?modelVersionId=2014522">lightx2v wan 2.1 i2v Rank64</a> accelerator lora with 0.68 strenght.- In the Advanved Ksamplers, use 4-6 total steps, CFG 1, with sampler Euler and Scheduler Beta57, doing half of the steps in each Ksampler (for example: 0 to 2 in High Noise Ksampler and 2 to 4 in Low Noise Ksampler with 4 total steps) - In the ModelSampling nodes use 15.87 shift (this is the optimal value for Beta57 scheduler).I tried all kind of methods (with 3 advanced ksamplers, using the accelerator lora only in low noise phase, etc...) and the one mentioned above is the one that works best, seen from a Reddit user comment some time ago. It gives the best quality in the shortest time avoiding slow motion videos for the most part. All the sample videos from the galery where generated this way and includes the workflow in the metadata.

Alfonso Azpiri Style Wan 2.2. t2v

The High Noise version of the LoRa (for people who prefer to have it in separate .safetensors files).

Files with SHA256: b4cb55fc6da5287d24787778f93a5b6b0e766e30a1757c11e4a7098dd0917599