Browse categories
Explore
Fiverr Pro
English
$
USD
I will train a custom character lora and deliver a consistent photo series
Fulvio G
About this gig
Your AI character changes face every time you generate a new image. Different lighting, different angle different person.
I fix that.
I train a custom LoRA for your character and deliver finished images where the identity actually holds: across lighting, angles, wardrobe and framing, from close-up to full body.
How it works:
- You send a brief, or reference images of an original character you own
- I build and curate the training dataset with automated identity scoring
- I train the LoRA on cloud GPUs and validate results against identity metrics
- I generate, select and upscale the final images
I'm a systems engineer with 15+ years in production infrastructure, and I built my own training pipeline instead of clicking through someone else's. That's why results are reproducible, not lucky.
Because the character is synthetic, there are no model releases, no expiring usage rights and no limits on reuse.
I only work on original or synthetic characters I don't train models on real people, public figures or celebrities.
Not sure which package fits? Message me before ordering.
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.
AI Avatar Design disclosure
I understand that by acquiring AI Avatar Design services, I must abide by any and all disclaimer rules required by applicable laws and the platform(s) where this content will be posted. Read more about Fiverr’s AI Guidelines.
Get to know Fulvio G
Fulvio G
AI Infrastructure and ComfyUI Engineer, Linux and LoRA training
- FromItaly
- Member sinceSep 2026
Languages
English, Italian
Linux systems engineer, 15+ years automating infrastructure for enterprise, healthcare and banking — virtualization, Terraform, Ansible, GitOps.
I now bring that engineering discipline to AI: ComfyUI pipelines deployed and optimised on local and cloud GPUs, and character LoRAs trained with a tested pipeline (automated dataset curation, ArcFace identity scoring, RunPod orchestration).
If you need a character that stays consistent across every shot — or an AI setup that runs reliably in production, not just on a laptop — that's what I do.
My Portfolio
FAQ
What counts as a revision?
Regenerating images from the trained model — different poses, lighting, framing or wardrobe. Changing the character's identity requires retraining the model, which is a new order. Tell me the look you want before I start training and we'll get it right the first time.
Can you train a model on a real person or a celebrity?
No. I only work with original or fully synthetic characters. This keeps both of us clear of likeness and image-rights problems.
Do I get the model file?
From the Standard package up, yes: you receive the trained .safetensors file and can keep generating on your own. You can also add it to the Basic package as an extra.
Can I use the images commercially?
Yes. The character is synthetic, so there are no model releases and no usage-rights limits on where or how long you use the images.
How consistent is "consistent"?
I validate identity retention with automated face-similarity scoring, not by eye. You'll see the same face across every shot in the set, in very different lighting.
