Fine tuning Dreambooth Flux Kontext I2I Error: the following arguments are required: --instance_prompt

[MichaelMelgarejoFlorez]

Describe the bug

Hello HF team, @sayakpaul @bghira

I'm encountering a persistent issue when trying to fine-tune the black-forest-labs/FLUX.1-Kontext-dev model using the train_dreambooth_lora_flux_kontext.py script.

I am following the official README instructions for Image-to-Image (I2I) finetuning. My goal is to train a transformation on my own dataset, which is structured for I2I (condition image, target image, and text instruction).

The Problem

Every time I run the script with the correct arguments for I2I finetuning, I get a : the following arguments are required: --instance_prompt

When I run this [Reproduction], I receive the error: the following arguments are required: --instance_prompt.

To isolate the issue from my personal dataset, I also tested the exact example command provided in the documentation (the one using kontext-community/relighting). I found that this command also fails with the identical the following arguments are required: --instance_prompt error.

Given that both my custom command and the official example command are failing in the same way, I am trying to understand the origin of this error. It seems the --instance_prompt argument is being required even when all I2I-specific arguments are provided.

Environment

Script: examples/dreambooth/train_dreambooth_lora_flux_kontext.py

Diffusers Version: I am using the specific commit 05e7a854d0a5661f5b433f6dd5954c224b104f0b (installed via pip install -e . from a clone), as recommended in the README.

Could you please help me understand why this might be happening? Is this expected behavior, or am I perhaps missing a configuration step?

Thank you for your time!

Reproduction

How to Reproduce

I am running the following command, which provides all the necessary arguments for I2I finetuning using my (dataset_name, image_column, cond_image_column, and caption_column) using my public dataset:

accelerate launch /local-git-path/train_dreambooth_lora_flux_kontext.py \
  --pretrained_model_name_or_path="black-forest-labs/FLUX.1-Kontext-dev" \
  --output_dir="/local-path/kontext-finetuning-v1" \
  --dataset_name="MichaelMelgarejoTotto/mi-dataset-kontext" \
  --image_column="output" \
  --cond_image_column="file_name" \
  --caption_column="instruccion" \
  --mixed_precision="bf16" \
  --resolution=1024 \
  --train_batch_size=1 \
  --guidance_scale=1 \
  --gradient_accumulation_steps=4 \
  --gradient_checkpointing \
  --optimizer="adamw" \
  --use_8bit_adam \
  --cache_latents \
  --learning_rate=1e-4 \
  --lr_scheduler="constant" \
  --lr_warmup_steps=200 \
  --max_train_steps=1000 \
  --rank=16 \
  --seed="0" 

Logs

train_dreambooth_lora_flux_kontext.py: error: the following arguments are required: --instance_prompt

System Info

• 🤗 Diffusers version: 0.35.0.dev0
• Platform: Linux-4.18.0-305.3.1.el8.x86_64-x86_64-with-glibc2.28
• Running on Google Colab?: No
• Python version: 3.10.19
• PyTorch version (GPU?): 2.7.1+cu118 (False)
• Flax version (CPU?/GPU?/TPU?): not installed (NA)
• Jax version: not installed
• JaxLib version: not installed
• Huggingface_hub version: 0.36.0
• Transformers version: 4.57.1
• Accelerate version: 1.11.0
• PEFT version: 0.17.1
• Bitsandbytes version: 0.48.1
• Safetensors version: 0.6.2
• xFormers version: not installed
• Accelerator: NA
• Using GPU in script?:
• Using distributed or parallel set-up in script?:

Who can help?

No response

[charming2992]

Also, the param "--cond_image_column=file_name" are not use in the code, which just training text2image

[asomoza]

Hi @MichaelMelgarejoFlorez , I did a quick test and that second example for i2i worked fine and started the training process, I also tried your command changing the script location and the output path and it also worked.

The recommendation to install from a specific commit probably was to ensure that this script and the kontext pipeline were present but after the release, it is better to just use the latest version or install from main. Can you install diffusers latest version or from main and test it?

if it works for you with this, probably we should remove the install instruction from a specific commit cc: @stevhliu @linoytsaban

[asomoza]

Also, the param "--cond_image_column=file_name" are not use in the code, which just training text2image

Hi @charming2992, it is used in several parts in the script, but probably the most meaningful one is:

diffusers/examples/dreambooth/train_dreambooth_lora_flux_kontext.py

Lines 834 to 837 in 500b9cf

	cond_image_column = args.cond_image_column
	if cond_image_column is not None:
	cond_images = [dataset["train"][i][cond_image_column] for i in range(len(dataset["train"]))]
	assert len(instance_images) == len(cond_images)

can you elaborate a bit on why do you think it's not used?

[charming2992]

Also, the param "--cond_image_column=file_name" are not use in the code, which just training text2image

Hi @charming2992, it is used in several parts in the script, but probably the most meaningful one is:

diffusers/examples/dreambooth/train_dreambooth_lora_flux_kontext.py

Lines 834 to 837 in 500b9cf

cond_image_column = args.cond_image_column
if cond_image_column is not None:
cond_images = [dataset["train"][i][cond_image_column] for i in range(len(dataset["train"]))]
assert len(instance_images) == len(cond_images)
can you elaborate a bit on why do you think it's not used?

Thanks for reply, I found it in the latest code.
I was following the training instructions for flux-kontext in README_flux.md, which requires switching the branch to 05e7a85. This indeed is not being used.

[sayakpaul]

Thanks @asomoza for the investigation! I think we can safely remove any specific commit info from the instructions.

[MichaelMelgarejoFlorez]

Hi @asomoza @sayakpaul

Thanks for the quick response. I've run my code using the main Diffusers branch and it worked perfectly. 👍

[sayakpaul]

Closing the issue then :)