Browse categories
Explore
Fiverr Pro
English
$
USD
I will train and test a custom h3 product style lora
About this Gig
I will train and test one H3 product-style LoRA using your approved reference images.
Please message me before ordering to agree on the style, reference rights, compatible H3 setup and test scenes.
Delivery includes:
- One H3-native .safetensors LoRA
- Trigger phrase, starting strength and tested ComfyUI workflow
- Six evaluation images: three matched base/LoRA pairs with prompts and seeds
- A report on observed changes, limitations and failed scenes
- One in-scope revision
This is a focused training pilot. Improvement in every image is not guaranteed. Exact product geometry, logos, lettering, identity and skin retouching are outside this package. SDXL/FLUX compatibility, hosting and installation are not included.
Training runs locally. Your files are not made public or reused for other clients without permission. You need suitable reference rights and access to the agreed H3 model; base weights and third-party assets are not supplied.
Messages in any language are welcome. I use translation tools when needed and confirm key requirements before starting.
Programming Language:
Python
AI Model Frameworks & Tools:
PyTorch
Data Type:
Images
AI Engine:
Other
My Portfolio
FAQ
What is included?
One custom H3 product-style LoRA, a trigger and starting strength, a tested ComfyUI workflow, three matched base/LoRA pairs (six images), a results report and one in-scope revision. This is one coherent style, not multiple identities or product lines.
Which model and setup are supported?
My current proof uses MiniMax H3 FL2VA in ComfyUI on Windows with an RTX 5090. I check your exact variant and node setup before accepting the brief. SDXL, FLUX, all H3 variants and installation on other computers are not automatically included.
Will it preserve my exact product or fix people?
No exact product geometry, logo, text, face identity or skin/anatomy correction is included. A style LoRA can change framing and shape. This package trains a product-photo look; it is not reference-faithful image retouching.
How many reference images do you need?
Send 12–30 clear, varied images for one coherent style, with training rights confirmed. I review them before the order and agree on the training and holdout split. Quantity alone does not establish quality; near-duplicates may be excluded.
What does one revision cover?
One focused adjustment, including up to one retraining pass if needed, within the agreed style, approved references and test brief. A new style, replacement dataset or new architecture is a new scope. Extra changes must be agreed on Fiverr first.
Can I use the model commercially?
Use depends on your H3 base-model permission, reference rights and intended purpose. I confirm the agreed scope before accepting the brief. Base-model weights and third-party assets are not included, and no blanket commercial-rights grant is promised.
How are my files handled?
Training and evaluation run locally after you submit files through Fiverr. I do not publish them, reuse them for other clients or send them to external model services without permission. Retention and deletion timing are agreed before the order.
Can I contact you in my language?
Messages in any language are welcome. I use translation tools when needed and confirm key requirements before starting. This does not mean native fluency in every language; the agreed brief is confirmed in writing on Fiverr.
What if the tests do not meet the agreed brief?
I report the limitation and use the included revision within scope. A failed scene is not presented as a quality improvement. Any unresolved issue is handled through Fiverr; this pilot does not promise a successful outcome for every prompt.
