Bring a Legacy Model LoRA Into the Modern Era
Some great LoRAs are still stuck in older ecosystems. This bounty is about giving them a new home.
Your challenge is to take a LoRA originally made for SD 1.5, SDXL, Flux.1, or one of their finetunes (Pony, Illustrious, NoobAI, etc.), update or rebuild its dataset, and retrain it for a newer model or models released from the HiDream era (April 2025) forward.
This is about more than just making pretty images. It is about preserving older work, learning newer model families, and keeping useful LoRAs relevant instead of letting them fade away.
Prizes
1st Place: 20,000 Buzz
2nd Place: 10,000 Buzz
3rd Place: 5,000 Buzz
The Goal
Start with an older LoRA and successfully migrate, adapt, or reinterpret it for a newer model family.
The best entries will show that you did not just make something new from scratch wearing an old name tag. You should preserve the core identity, concept, subject, or stylistic intent of the original while making it perform well on a newer model.
Eligible Starting Points
You may begin with:
A legacy LoRA that you originally created
A legacy LoRA from an inactive or abandoned account, with clear attribution to the original creator.
Required Submission Materials
Your entry must include all of the following:
A link to the original LoRA
A short note explaining whether it was your own LoRA or one you are reviving with attribution
The new LoRA, published to your Civitai account
example images made with the new LoRA to the bounty
A writeup covering:
what you kept from the original
what newer model you trained for
any important lessons, challenges, or improvements
Rules
Your source LoRA must come from an older ecosystem such as SD 1.5, SDXL, Flux.1, or their finetunes (Pony Illustrious, NoobAI, etc.).
Your new version must be trained for a newer model or models from the HiDream era forward (April 2025-Present).
The entry must show a meaningful relationship to the original LoRA’s identity. Same character, same subject, same concept, same visual theme, same style intent, or a clearly faithful reinterpretation all count.
A completely unrelated retrain does not qualify.
You must publish the new LoRA to your account as part of the submission.
If you are using someone else’s older LoRA, you must provide clear attribution.
All entries must comply with Civitai Terms of Service.
Entry Notes and Civitai Articles are strongly recommended.
Bounty images limited to R rating (Do what you will with the LoRA and datasets)
Judging Criteria
Entries will be judged on:
Faithfulness to the original
Does it still feel like the same LoRA, concept, subject, or style?Quality on the newer model
Does it actually perform well in its new ecosystem?Showcase images
Do the example images clearly demonstrate the strength of the migration?
What I’m Looking For
I want to see entries that treat this like a real move, not just a quick repaint.
Show me:
old ideas brought forward well
forgotten LoRAs made relevant again
great example outputs that prove it works
In the Spirit of the Bounty
This challenge is about preservation, modernization, and learning newer models.
Some LoRAs deserve a second life. Some deserve their first real upgrade. Pack the boxes, haul the weights, and bring something worth keeping into the modern era.
Good luck, and happy moving. 📦✨