Team
Shared LoRAs
Train one in a team workspace and it belongs to the team, usable by everyone and absent from every personal library including yours.
Train a LoRA while working in a team workspace and it belongs to the organisation. Every member can see it and generate with it, from any board in that workspace, without anybody sharing anything. It is how a trained look becomes house style without anybody having to distribute a file. ## The rule, precisely | Trained in | Belongs to | Appears in | |---|---|---| | A team workspace | The organisation | The team library, for every member | | Your personal workspace | You | Your personal library only | A team LoRA appears in **no personal library at all**, including the library of the person who trained it. It is not copied to the team and kept privately as well. :::note[Which workspace you were in decides ownership] Not who you are, not who paid, and not which board. Training while switched into the team workspace produces a team LoRA; training in your personal workspace produces a personal one. There is no reassignment afterwards, so switch before you train. ::: ## A style guide that produces the look A written brand guideline describes a look and depends on everyone reading it, interpreting it the same way, and remembering it under deadline. A shared LoRA produces the look. A new member generates on-brand on their first afternoon without having read anything, because the adapter carries what the document would have described. That is worth building deliberately: - **A product LoRA** per thing you make repeatedly, so it is recognisably itself across every campaign - **A house style LoRA** trained on work that already went out, so new work matches the back catalogue - **A 3D style LoRA** trained on ten to twenty of your own textured models, so generated geometry arrives in your visual language ## Using one somebody else trained Exactly like using your own. Pick it in **lora-generate**, set its scale, and put its trigger word in the prompt. The trigger is the one thing that has to travel by word of mouth, so keep the LoRA's name and its trigger obviously related. A team LoRA named "Aurora chair" with a trigger nobody can guess is a LoRA nobody uses. :::tip[Write the trigger into a sticky note on the team's chains board] The same board that holds the sequences the team relies on is the right home for the triggers. It costs one note and it is the difference between a trained look being used and being rediscovered every few months. ::: ## Stacking a house style with a subject Run the product LoRA high and the house style LoRA low, and your object arrives rendered the way your studio renders things. The one hard constraint is generational. FLUX.1 and FLUX.2 adapters run through different endpoints and cannot be combined in a single run, so if you intend to stack, train both halves on the same generation.Train a LoRA while working in a team workspace and it belongs to the organisation. Every member can see it and generate with it, from any board in that workspace, without anybody sharing anything.
It is how a trained look becomes house style without anybody having to distribute a file.
The rule, precisely
Section titled “The rule, precisely”| Trained in | Belongs to | Appears in |
|---|---|---|
| A team workspace | The organisation | The team library, for every member |
| Your personal workspace | You | Your personal library only |
A team LoRA appears in no personal library at all, including the library of the person who trained it. It is not copied to the team and kept privately as well.
A style guide that produces the look
Section titled “A style guide that produces the look”A written brand guideline describes a look and depends on everyone reading it, interpreting it the same way, and remembering it under deadline.
A shared LoRA produces the look. A new member generates on-brand on their first afternoon without having read anything, because the adapter carries what the document would have described.
That is worth building deliberately:
- A product LoRA per thing you make repeatedly, so it is recognisably itself across every campaign
- A house style LoRA trained on work that already went out, so new work matches the back catalogue
- A 3D style LoRA trained on ten to twenty of your own textured models, so generated geometry arrives in your visual language
Using one somebody else trained
Section titled “Using one somebody else trained”Exactly like using your own. Pick it in lora-generate, set its scale, and put its trigger word in the prompt.
The trigger is the one thing that has to travel by word of mouth, so keep the LoRA’s name and its trigger obviously related. A team LoRA named “Aurora chair” with a trigger nobody can guess is a LoRA nobody uses.
Stacking a house style with a subject
Section titled “Stacking a house style with a subject”Run the product LoRA high and the house style LoRA low, and your object arrives rendered the way your studio renders things.
The one hard constraint is generational. FLUX.1 and FLUX.2 adapters run through different endpoints and cannot be combined in a single run, so if you intend to stack, train both halves on the same generation.