[bug]: Flux Kontext with 2/3 reference images - Out of Memory Error (low VRAM 4GB)

[xen0form]

Is there an existing issue for this problem?

• I have searched the existing issues

Install method

Invoke's Launcher

Operating system

Linux

GPU vendor

Nvidia (CUDA)

GPU model

RTX 3050 Laptop

GPU VRAM

4

Version number

6.3.0

Browser

No response

System Information

No response

What happened

I have enabled the recommended "enable_partial_loading: true", tried enabling "pytorch_cuda_alloc_conf: "backend:cudaMallocAsync"" and setting device_working_mem_gb: 3", nothing helps, I keep getting OOM errors when using more than 3 reference images, sometimes (randomly) when only 2 of them.

What you expected to happen

I was hoping the enable_partial_loading option would help with offloading VRAM, as it had previously worked perfectly, so i'll be able to use more than 2 reference images for the Flux Kontext.

How to reproduce the problem

No response

Additional context

No response

Discord username

No response

[xen0form]

torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 762.00 MiB. GPU 0 has a total capacity of 4.00 GiB of which 0 bytes is free. Of the allocated memory 3.15 GiB is allocated by PyTorch, and 146.45 MiB is reserved by PyTorch but unallocated.

If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)

[psychedelicious]

Please paste your system information as requested in the GH issue template:

Click the gear icon at the bottom left corner, then click "About". Click the copy button and then paste here.

[xen0form]

{
"version": "6.3.0",
"dependencies": {
"absl-py" : "2.3.1",
"accelerate" : "1.8.1",
"annotated-types" : "0.7.0",
"anyio" : "4.9.0",
"attrs" : "25.3.0",
"bidict" : "0.23.1",
"bitsandbytes" : "0.46.1",
"blake3" : "1.0.5",
"certifi" : "2025.7.9",
"cffi" : "1.17.1",
"charset-normalizer" : "3.4.2",
"click" : "8.2.1",
"colorama" : "0.4.6",
"coloredlogs" : "15.0.1",
"compel" : "2.1.1",
"contourpy" : "1.3.2",
"CUDA" : "12.8",
"cycler" : "0.12.1",
"Deprecated" : "1.2.18",
"diffusers" : "0.33.0",
"dnspython" : "2.7.0",
"dynamicprompts" : "0.31.0",
"einops" : "0.8.1",
"exceptiongroup" : "1.3.0",
"fastapi" : "0.116.1",
"fastapi-events" : "0.12.2",
"filelock" : "3.18.0",
"flatbuffers" : "25.2.10",
"fonttools" : "4.58.5",
"fsspec" : "2025.5.1",
"gguf" : "0.17.1",
"h11" : "0.16.0",
"httptools" : "0.6.4",
"huggingface-hub" : "0.33.4",
"humanfriendly" : "10.0",
"idna" : "3.10",
"importlib_metadata" : "8.7.0",
"invisible-watermark" : "0.2.0",
"InvokeAI" : "6.3.0",
"jax" : "0.6.2",
"jaxlib" : "0.6.2",
"Jinja2" : "3.1.6",
"kiwisolver" : "1.4.8",
"markdown-it-py" : "3.0.0",
"MarkupSafe" : "3.0.2",
"matplotlib" : "3.10.3",
"mdurl" : "0.1.2",
"mediapipe" : "0.10.14",
"ml_dtypes" : "0.5.1",
"mpmath" : "1.3.0",
"networkx" : "3.4.2",
"numpy" : "1.26.4",
"onnx" : "1.16.1",
"onnxruntime" : "1.19.2",
"opencv-contrib-python": "4.11.0.86",
"opencv-python" : "4.11.0.86",
"opt_einsum" : "3.4.0",
"packaging" : "25.0",
"picklescan" : "0.0.26",
"pillow" : "11.3.0",
"pip" : "25.1.1",
"prompt_toolkit" : "3.0.51",
"protobuf" : "4.25.8",
"psutil" : "7.0.0",
"pycparser" : "2.22",
"pydantic" : "2.11.7",
"pydantic-settings" : "2.10.1",
"pydantic_core" : "2.33.2",
"Pygments" : "2.19.2",
"pyparsing" : "3.2.3",
"PyPatchMatch" : "1.0.2",
"pyreadline3" : "3.5.4",
"python-dateutil" : "2.9.0.post0",
"python-dotenv" : "1.1.1",
"python-engineio" : "4.12.2",
"python-multipart" : "0.0.20",
"python-socketio" : "5.13.0",
"PyWavelets" : "1.8.0",
"PyYAML" : "6.0.2",
"regex" : "2024.11.6",
"requests" : "2.32.4",
"rich" : "14.0.0",
"safetensors" : "0.5.3",
"scipy" : "1.15.3",
"semver" : "3.0.4",
"sentencepiece" : "0.2.0",
"setuptools" : "80.9.0",
"simple-websocket" : "1.1.0",
"six" : "1.17.0",
"sniffio" : "1.3.1",
"sounddevice" : "0.5.2",
"spandrel" : "0.4.1",
"starlette" : "0.47.1",
"sympy" : "1.14.0",
"tokenizers" : "0.21.2",
"torch" : "2.7.0+cu128",
"torchaudio" : "2.7.0+cu128",
"torchsde" : "0.2.6",
"torchvision" : "0.22.0+cu128",
"tqdm" : "4.67.1",
"trampoline" : "0.1.2",
"transformers" : "4.53.2",
"typing-inspection" : "0.4.1",
"typing_extensions" : "4.14.1",
"urllib3" : "2.5.0",
"uvicorn" : "0.35.0",
"watchfiles" : "1.1.0",
"wcwidth" : "0.2.13",
"websockets" : "15.0.1",
"wrapt" : "1.17.2",
"wsproto" : "1.2.0",
"xformers" : "0.0.30",
"zipp" : "3.23.0"
},
"config": {
"schema_version": "4.0.2",
"legacy_models_yaml_path": null,
"host": "127.0.0.1",
"port": 9090,
"allow_origins": [],
"allow_credentials": true,
"allow_methods": [""],
"allow_headers": [""],
"ssl_certfile": null,
"ssl_keyfile": null,
"log_tokenization": false,
"patchmatch": true,
"models_dir": "models",
"convert_cache_dir": "models\.convert_cache",
"download_cache_dir": "models\.download_cache",
"legacy_conf_dir": "configs",
"db_dir": "databases",
"outputs_dir": "outputs",
"custom_nodes_dir": "nodes",
"style_presets_dir": "style_presets",
"workflow_thumbnails_dir": "workflow_thumbnails",
"log_handlers": ["console"],
"log_format": "color",
"log_level": "info",
"log_sql": false,
"log_level_network": "warning",
"use_memory_db": false,
"dev_reload": false,
"profile_graphs": false,
"profile_prefix": null,
"profiles_dir": "profiles",
"max_cache_ram_gb": null,
"max_cache_vram_gb": null,
"log_memory_usage": false,
"device_working_mem_gb": 0.8,
"enable_partial_loading": true,
"keep_ram_copy_of_weights": true,
"ram": null,
"vram": null,
"lazy_offload": true,
"pytorch_cuda_alloc_conf": null,
"device": "auto",
"precision": "auto",
"sequential_guidance": false,
"attention_type": "auto",
"attention_slice_size": "auto",
"force_tiled_decode": false,
"pil_compress_level": 1,
"max_queue_size": 10000,
"clear_queue_on_startup": false,
"allow_nodes": null,
"deny_nodes": null,
"node_cache_size": 512,
"hashing_algorithm": "blake3_single",
"remote_api_tokens": null,
"scan_models_on_startup": false,
"unsafe_disable_picklescan": false
},
"set_config_fields": ["enable_partial_loading", "legacy_models_yaml_path", "device_working_mem_gb"]
}

[psychedelicious]

Thanks. It looks like the OOM is occuring during VAE Decode (last step in generation process). This is often the most memory-intensive step. The amount of memory required is dependent on the image dimensions. I wouldn't expect the number of kontext ref images to have an effect.

How large are the images you are generating?

[xen0form]

OOM occurs at the beginning of the generation, not VAE. It didn't even start generating.
Output (and each of 3 Kontext references) image size 832 x 1248.

[psychedelicious]

Ah I see, thanks. It's OOMing during VAE encode (FLUX Kontext needs encode the ref images using the VAE).

I improved the memory management for VAE encoding in #8414 but I don't think it will help your case. My tests show that with 4GB of VRAM, you are right on the edge of being able to encode two ref images at that size.

We'd need to use a different strategy when preparing Kontext ref images to avoid the VRAM spike you are hitting (encode first, then concat in latent space instead of the other way around).

[psychedelicious]

I implemented the other strategy in that PR. I don't think it will let you use 3 ref images reliably, but 2 should work reliably now.