A training run failed
Last updated: October 15, 2026
A finetune did not train: the run shows Failed, it has been running far longer than usual, or the Create finetune button will not start it. Find what you see below.
Where to read the reason
Open the reference set. Under Finetunes, the model's row reads Failed with Open to see why. Open it: a Last training failed banner above the references gives the reason and a button for the fix, and View run opens the run in Runs. You also get an email when a run fails.
A finetune trained earlier on the same model keeps working while you fix the problem and retry.
What happens to your Creative Units
Creative Units are held when a run starts and charged only when it finishes. A run that fails is not charged, and the units held for it are released. See Creative Units.
The run failed
Training took too long and timed out
The banner reads Training took too long and timed out. Try again, or reduce the number of images or steps. Runs have a time limit, and a large set with many epochs can pass it.
Fix: click Adjust settings. The training form opens its advanced settings and lowers Epochs to about halfway between what timed out and the minimum, marking the old value so you can see the change. Train again. If the set is very large, removing near-duplicate images also shortens the run.
Some training inputs couldn't be used
The banner names the input that was rejected, or reads Some training inputs couldn't be used. Check that every image opens and that edit pairs have both a before and an after, then try again.
Fix: click Review pairs to jump to the set's pairs and files. Replace any file that does not open, give every edit pair its before and after image and an edit description, then train again.
Not enough Creative Units
The banner reads Not enough creative units to run this training. The workspace balance ran short before training could start.
Fix: click Top up, add units, and train again. We couldn't verify your workspace balance has the same button; it usually clears if you try again shortly.
The training provider couldn't finish this run
The banner shows the provider's own message, or The training provider couldn't finish this run. The service that runs the training refused or dropped the job.
Fix: if the message names something in your set (a file, a setting), change it. Otherwise try again later. If it keeps failing, click Get help.
Something went wrong on our end
The banner reads Something went wrong on our end during training, or says the reason is unknown. Try again; if it fails a second time, click Get help and contact support.
The run seems stuck
Taking longer than expected
While a run trains, its banner shows progress and the time left. When a run has gone a long while with no estimate, the banner changes to Taking longer than expected and says the run may have stalled. A run that reads Finalizing… is saving the finetune and rendering previews; give it time.
Fix: click Stop training and confirm, then start the run again. Stopping produces no finetune. If runs stall repeatedly on the same set, contact support.
Training won't start
When Create finetune is greyed out, hover it to read why.
Add at least 6 files to train
An image finetune needs a minimum number of images. Add more to the set. Only image files count towards an image finetune.
Add more images to train
Automatic cannot pick a model and settings for the files the set has. Add images, or switch to Customize and pick a model yourself.
Give at least one pair a before image to train
An edit finetune learns from before-and-after pairs, and none of the pairs has its before image yet. Add one, or use Generate before on the pair. See Train an edit finetune from before/after pairs.
Describe the edit on N pairs to train
Every pair in an edit finetune needs a description of the change. The message counts the pairs without one. Fill in Describe the edit on each, or write one and click Apply to all pairs beside it.
No compatible base model is available for this workspace
Automatic found no model it may train for this workspace. Switch to Customize and pick a model from the list, or ask a workspace admin which models the workspace allows.
Not enough creative units to train this model
The button reads Not enough creative units to train this model. Click to top up. Click it to add units, then start the run.
Cannot run training, as you have exceeded
Each person can start a limited number of training runs per hour. Wait a while and try again, or ask a teammate with the Train styles permission to start it.
Start another training run?
Not an error: a run is already training for this model, and confirming starts a second one alongside it that spends Creative Units again. Cancel if you meant to wait for the first.
Still stuck?
Contact support from Help & Resources > Talk To Us in Layer. Include the reference set's name, the model you trained, the time of the run and the message on the Last training failed banner. Contact support with the right details lists everything that helps.