Version 2 ReduxOkay, sorry for the drama but my 2.0 model is back up because the sampler I was using avoid the plastic skin look was also creating weird noise bands with all my renders. Use whatever Scheduler/Sampler you want to use but I do suggest use this LoRa at a 0.8 strength. Thank you for your patience! I've also added more photos of how this LoRa works with another style LoRA like Midjourney Style. Enjoy more Chillenhaal!<s>Blake Chillenhaal is back and better than ever in v2.0. I've decided to add more full body shots of Jake in his dataset because in v1, the focus was more on headshots, thus disfigured body proportions. I hope some of the over baking was fixed too in v2.0. I'm now on runpod, so I can create better renders on fp8 flux.</s><s>Recommended Settings for fp8 Flux:</s><ul><li><s>Scheduler: sgn_uniform</s></li><li><s>Steps: 35</s></li><li><s>Sampler: deis (avoids the plastic skin look)</s></li><li><s>Flux Guidance: 4.0</s></li><li><s>LoRA Strength: 0.9</s></li></ul><s>Version 2 was a disaster. Much worse than v1, so please, if you downloaded v2 make sure you stick with v1 for your Chillenhaal needs. I'm going to tinker what happened and why I'm getting distorted results.</s><hr />Version 1Who wouldn't like a Blake Chillenhaal in your LoRA folder? This model was created using Flux <a target="_blank" rel="ugc" href="http://1.Dev">1.Dev</a> but the showcase images were created on Flux Schnell, because I'm on an Apple M2 Macbook Pro with NO graphics card. I also used Unlimited SD Upscaler at the denose level of 0.15. I'll be updating this model as I get better at training LoRA.If you're having difficulty with Jake's face/likeness showing up when pairing with other LoRAs, just remember that other LoRAs are trained on faces too and you'll have to play around with block weights to find the face layer or just tweak the strength levels.Recommended Scheduler: sgm_uniformUpdate: I've uploaded a few renders using Flux 1.D!<h1 id="resource-shoutouts-15q5tfbsk">Resource Shoutouts</h1>I used <a target="_blank" rel="ugc" href="https://github.com/ssitu/ComfyUI_UltimateSDUpscale">Ultimate SD Upscaler</a> using the "NMKD-Siax_200k" model at denoise level set between 0.15-0.20 for the renders that used Flux Schnell.I also used that awesome <a target="_blank" rel="ugc" href="https://civarchive.com/models/686449/star-trek-tng-uniforms-3-variants-flux">Star Trek TNG Uniforms Flux LoRa</a> is is only trained on the uniforms which doesn't mess with the face as much.I also used <a target="_blank" rel="ugc" href="https://civarchive.com/models/551747/vaporwave-sdxl-pony-flux">VaporWave Flux LoRa</a> for the miami vice style image though you'll have to really set the strength level down because it'll override your character LoRA if you have one set.

Jake Gyllenhaal - Flux 1.D

v1.0 - Trained on 13 images, trains per image = 10, epochs = 16, network dim = 4

Files with SHA256: 6d773183a44b1539230ce0b44708d3724f9fd097cf1085c8336eec37c5366bd0