<h2 id="heading-16">My rating vs expectation: 90%</h2>(Very high quality dataset. Angles of the photos could be improved, which led to less than great likeness. Or perhaps it's still under-trained even at 472 epochs. Bummer it didn't turn out perfect. Also it's highly baked, so not very flexible with simple prompts.)All TIs I make depict real adults. Please apply it with good judgement. If you're the model and want this removed reach out to me via link below. I welcome all criticisms and suggestions with curiosity. If you like my work, buy me a house: <a target="_blank" rel="ugc" href="https://ko-fi.com/futurafree">https://ko-fi.com/futurafree</a>Experimenting with high step count, larger dataset and low learning rate and I'm having a bit more success with it. This TI was trained in sd1.5 base with 5 vectors, 52 images, 24525 steps, and 472 epochs. This was finetuned for epiCRealism evo3 and Realistic Vision V4.0, but may work on others. TIPS:<ul><li>MUST be the first word in the prompt for likeness and consistency.<ul><li>Example: futurahannah24525v5, portrait of a woman, long dress etc.</li><li>Not: Portrait of futurahannah24525v5, etc.</li></ul></li><li>Recommended strength:<ul><li>(0.98-1.0)</li></ul></li></ul><ul><li>I use adetailer for all images, with mediapipe_face_full, blank pos and neg, 1024x1024, inpaint mask blur: 20, inpaint noise strength: 0.45.</li><li>Again, prompt in adetailer is left blank most of the time; (*exception*: if Loras/Lycoris is used and has significant impact on the appearance of the model's face, copy paste pos prompt without Loras/Lycoris into adetailer).</li><li>Upscaler I use: 4x_NMKD-Siax_200k.</li></ul>

Files with SHA256: 6361338e1b5c788481d8e64c865cf5cef8b0e133968bd0249ae9857bf7f14e1c