<h3 id="are-you-ready-to-perform-your-first-human-transmutation-with-your-favorite-character">Are you ready to perform your first human transmutation with your favorite character?</h3><h3 id="just-be-careful...-there-is-always-a-price-to-pay.">Just be careful… there is always a price to pay. 😁</h3>This LoRA is inspired by the Gate of Truth from Fullmetal Alchemist.Use it for intense, mysterious, or slightly cursed transmutation scenes, but don’t blame me if Truth decides to take something in return. <h3 id="lora-strength">LoRA Strength</h3> I would not recommend moving too far away from a strength value of 1.0, since that seems to be the most reliable balance for this LoRA. As a general reference: Model strength: around 1.0 CLIP strength: around 0.75–1 You can experiment a little depending on the scene, but staying close to 1.0 usually gives the best results.I also noticed that the LoRA is more stable with my own checkpoint. I am currently testing it on other checkpoints, and in some cases the floor may not remain completely white. It can sometimes take on a different appearance, such as tiles, sand, or other ground textures.<h2 id="recommended-settings"> Recommended Settings</h2> For the sampler and scheduler, I usually recommend: Sampler: Euler Ancestral Scheduler: Simple Steps: 25–30 CFG: around 5 These settings worked well for me, but feel free to experiment with your own setup. <h3 id="prompting-tips">Prompting Tips</h3> I recommend not changing the keyword order too much, especially at the beginning of the prompt. It is important to keep: white background, white floor near the start of the prompt, because the LoRA can sometimes generate an unwanted floor or ground structure. I will try to improve this in a future update. Upscaling is often useful, especially for more detailed results. I recommend starting from a fairly large vertical resolution, such as: 1024x1536 This helps preserve more detail from the beginning while keeping a good vertical composition. <h3 id="future-updates">Future Updates</h3> Some of you asked me privately how I created this kind of image, so I decided to make a LoRA for it. I also thought this was an interesting concept to train, and it gave me a good opportunity to practice and improve my LoRA training workflow. I am still trying to improve my training skills so I can create more flexible and useful LoRAs in the future. For future updates, I may add some of the best images created by the community to improve the dataset and make the concept even more stable. I hope you enjoy using it. Try it with your favorite characters and show me what kind of forbidden transmutation scenes you can create.

Files with SHA256: 2d7a884eadd00ef26450ed301d1c4adc7d3d16493d086d2139961d4bdc8c685f