Ver categorias
Explorar
Fiverr Pro
Português
$
USD
I will train a custom lora for your character, mascot, or product
Jason R
Sobre este Serviço
I train a custom LoRA that teaches an image model your character, mascot, product, or art style, so it comes out consistent instead of almost right.
You send 15 to 30 images you own the rights to. I clean and caption the dataset, train on a commercially licensed base model (SDXL, FLUX.1 schnell, Qwen-Image, or Z-Image), and test the result before I hand it over. Every tier includes test renders, a trigger word, and the weight range that works best.
Standard adds a weight-sweep grid and a pack of 10 proven prompts. Premium is the full identity package: the LoRA, a ready-to-run ComfyUI workflow, a plain-English usage guide, and an eval sheet that runs 12 held-out prompts against the base model, so you can see exactly what the LoRA adds.
One rule up front: I only train on bases whose licenses allow commercial use, so you can legally sell work made with your LoRA. That is why I do not offer FLUX.1 dev. Its license does not allow selling derivatives. Fictional characters, products, and styles you own only. No real-person likenesses.
Conheça mais sobre Jason R
Jason R
- A partir deEstados Unidos
- Membro desdejan. de 2025
Idiomas
Inglês
Looking to enhance the quality of your processes, products, or services? With extensive experience in aviation quality, I bring a sharp, detail-focused approach to help businesses across industries. I can assist with process optimization, workflow improvements, and consistency checks in manufacturing, construction, retail, hospitality, energy, and education. Let’s work together to streamline operations, improve results, and deliver exceptional quality every time!
Perguntas frequentes
Which base model should I pick?
SDXL is the safe all-rounder with the biggest tool support. FLUX.1 schnell and Qwen-Image give stronger detail and follow prompts better. Z-Image is fast and sharp for stylized work. If you are not sure, pick "choose for me" and I will match the base to your subject and how you plan to use it.
Why don't you offer FLUX.1 dev? Everyone else does.
Because its license does not allow commercial use of derivatives, and a LoRA is a derivative. Plenty of sellers ignore that. I would rather sell you a LoRA you can actually use in paid work. The bases I train on (SDXL, FLUX.1 schnell, Qwen-Image, Z-Image) all allow commercial use.
How many images do you need?
For a character or product, 15 to 30 good images, varied in angle and setting, beat 100 near-duplicates. For an art style, send 30 or more. If your set is too small or too repetitive to train well, I will tell you before I start, not after.
Can you train a LoRA of me, or of a celebrity or influencer?
No. I train fictional characters, mascots, products, and art styles that you own or created. No real-person likenesses, mine included as a hard rule, not a price tier.
What exactly do I receive?
The LoRA as a.safetensors file, its trigger word, the recommended weight range, test renders, and settings notes, delivered on Fiverr. Standard adds the weight-sweep grid and prompt pack. Premium adds the ComfyUI workflow, the usage guide, the eval sheet, and one free retrain within 30 days.
What if the first training run is not right?
Training involves trial, which is why testing is built into every tier. Revisions cover picking a different checkpoint or adjusting the weight. Premium also includes one full retrain. If your images cannot support the subject, I will say so up front, and you can cancel before training starts.
Can I sell images I make with my LoRA?
Yes. That is the point of the licensed-base rule: the bases I use permit commercial use, so art and products you make with your LoRA are yours to sell. Your images stay yours too. I do not reuse buyer datasets or deliver your LoRA to anyone else.
