A request of @horsegirl.
Jenna Ortega is an American actress of Mexican and Puerto Rican descent. She rose into stardom with her role as Wednesday Addams in the Netflix show Wednesday, but she has also made appearences in films such as Scream VI or The Fallout.
This is a 135-step TI trained on a dataset of 15 images with these settings.
Curious about my work process? I have summarized it here.
Do you have a specific idea for a TI in mind? Visit my website and let me know.
Building a good prompt with my TIs
You're obviously free to experiment, but bear in mind that my TIs are trained with a more or less fixed phrasing, that normally starts with:
"photo of EMBEDDING_NAME, a woman"
So I recommend always starting your prompt like that and then building the rest of the prompt from there. For instance, "photo of beautiful (jenn0rtega-135:0.99), a woman with beautiful hair, as a movie star in a (movie premiere), premiere gala, (near a movie theatre), natural skin texture, (sexy tight {red|yellow|blue|green|black|turquoise} mesh strapless dress), 24mm, 4k textures, soft cinematic light, adobe lightroom, photolab, hdr, intricate, elegant, highly detailed, sharp focus, ((((cinematic look)))), soothing tones, insane details, intricate details, hyperdetailed, low contrast, soft cinematic light, exposure blend, hdr, faded, (paparazzi in background), (painted lips:1.1), ((looking at viewer:1.1))"
Please also note that I'm using the "add detail" LoRA for my example pics. I recommend setting it around 0.5 for best results.
Description
135-step TI trained on a dataset of 15 images with these settings.
Note: Retrained with 6 new pics in the dataset, replacing 6 other I judged not optimal. I wasn't completely happy with how the previous version captured Jenna's looks, especially when smiling, and I think this new version addresses that issue.
FAQ
Comments (6)
I'd love to your generation of Sadie Sink! It seems like many of them don't quite look like her
"A request of @horsegirl."
I totally read that as "a request of A horsegirl", like Jenna was a horse. :D
THIS is THE BEST model.
I have downloaded and used all the other 11 models ever created for the SD-1.4 and 1.5 and
this is THE BEST of all of them.
Thank you once again JG !
Thank you, Max! I hope this update will change your mind about which one is the best. ^^
I'm just doing TI right now. It's such a weird thing... My SDXL NEEDS the word "man" right after the embedding, like a trigger word. But SD15 does not. Same dataset, but with some different parameters.
Also, the weirdest thing is, it works with kind of the same with strength 0, or even negative strength... but not the SD15 version.
Mind I ask you, how many vectors did you use? Some people recommend 2 or 3... I don't know. My SDXL one achieve likeness with 3. But I need to do more testing.
Also, did you mean 150 steps per image? Or on total? Because if said you did 400 epochs... that's probably a lot more steps right?
Hey! Yeah, SDXL is still a mystery to me, I've never even tried it yet. :D Regarding your other questions, I've been using 4 vectors for some time now. Before that I used 2, and before that I basically tried different settings. Ultimately, I think I've obtained best results with 4 vectors, so that's why I stuck to that number. In principle, you shouldn't need more than 2, though. There's quite a lot of literature and many different views on the matter.
About steps, I always post all of the generation data for my TIs. My settings are usually as follows:
- Batch size: 3
- Gradient accumulation steps: 15
- 15 images on the dataset
With these settings, I run my training for a maximum of 140 steps usually (rarely need to go over that). Since I'm running 3 full epochs per step (batch size 3 * gradient 15 = 45, which is 3 times the number of images in my dataset), that means 140 steps equal 420 epochs.
I know the maths are a bit fuzzy, but I hope this makes some sense to you. Good luck with your training! :)
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.