Browse categories
Explore
Fiverr Pro
English
$
USD
I will train a custom lora model and build a comfyui workflow
AurixNotion
About this gig
I train custom LoRA models that reliably reproduce a specific character, product, or style, and build ComfyUI workflows that turn them into consistent images and videos.
Full process: dataset, captioning, training, and regularization, so the model stays accurate without leaking into unrelated generations.
No dataset? I'll build one from your raw photos or videos, or generate an original synthetic set for new characters and styles.
Workflows built around your LoRA:
- img2img: restyle and edit images
- img2vid: animate your character or product
- ref2vid: video from your reference images
- First/last frame to video: shots between two keyframes
- Anything else: describe the task, I'll tell you if it's doable
Models: SDXL and FLUX.2 [klein] 4B, both allow commercial use. Only images you have rights to; real people only with their consent.
Every LoRA is tested on new scenes and prompts, not just the training photos.
Premium: the workflow plus a one-click install script. If anything breaks, I'll guide your setup on a call (Windows + NVIDIA).
Send your references and goals, and I'll tell you honestly what's realistic before you order.
Delivery style preference
Please inform the freelancer of any preferences or concerns regarding the use of AI tools in the completion and/or delivery of your order.
Get to know AurixNotion
AurixNotion
Generative AI Specialist, ComfyUI and LoRA Training
- FromUkraine
- Member sinceSep 2026
- Avg. response time1 hour
Languages
English, Ukrainian, Russian
I train custom LoRA models and build ComfyUI workflows from scratch for image and video generation.
What I do:
- Custom LoRA training for products, characters, and styles (SDXL, FLUX.2, Z-Image)
- ComfyUI workflow design, built to your needs
- AI video and image-to-video (Wan, HunyuanVideo, MiniMax)
- Dataset prep, captioning, and regularization for reliable results
I run everything locally on my own workstation, so no cloud queues and fast iteration until the result is right.
Send me your reference material and I'll tell you honestly what's achievable.
FAQ
What images do I need to provide?
For a person, character, or object — at least 15-20 clear, high-resolution photos from different angles, lighting, and backgrounds. For a style — around 30-50 example images. More variety in your photos means a more flexible, reliable LoRA.
Which platform will I be able to use the LoRA on?
You'll get a standard .safetensors file that works in ComfyUI, Automatic1111, and most other Stable Diffusion / FLUX tools. If you need help setting it up, the Premium package includes a ready-to-use ComfyUI workflow.
How long does training actually take?
he training itself usually takes 1-3 hours depending on the model. Most of the delivery time goes into dataset preparation, captioning, and generating clean sample results — not the training run itself.
Will the LoRA work in scenes different from my training photos?
Yes, that's the goal. I test every LoRA with prompts describing new environments, lighting, and actions your training photos don't include, to make sure it generalizes rather than just repeating the training images.
Can I use the LoRA commercially?
Yes. I train on Apache 2.0 licensed base models (SDXL, FLUX.2 Klein) so there are no restrictions on commercial use of what you generate.
What if the first result isn't what I expected?
That's what revisions are for — included in every package. Send specific feedback (what's off, which sample) and I'll retrain or adjust accordingly.
